id="p-0050" num="0069">In some embodiments, the sample is obtained from a human subject. In some embodiments, the subject is suspected of having cancer. In some embodiments, the subject is suspected of having a viral infection. In some embodiments, the subject is suspected of having come into contact with a virus, such as smallpox virus, dengue virus, or human rhino virus.

In some embodiments, the toehold riboregulator comprises a nucleotide sequence of any one of SEQ ID NOs: 43841, 9602, 62866, and 19367.

In some embodiments, the toehold riboregulator comprises a nucleotide sequence of SEQ ID NO: 43841. In some embodiments, the toehold riboregulator comprises a nucleotide sequence of SEQ ID NO: 9602. In some embodiments, the toehold riboregulator comprises a nucleotide sequence of SEQ ID NO: 62866. In some embodiments, the toehold riboregulator comprises a nucleotide sequence of SEQ ID NO: 19367. In some embodiments, the toehold riboregulator comprises a toehold riboregulator comprising a nucleotide sequence of SEQ ID NO: 43841 and a toehold riboregulator comprising a nucleotide sequence of SEQ ID NO: 62866.

In some embodiments, the subject is suspected of having a smallpox virus infection or having been exposed to smallpox virus. In some embodiments, the subject is suspected of having dengue virus infection or having been exposed to dengue virus. In some embodiments, the subject is suspect of having human rhino virus infection or having been exposed to human rhino virus.

In some embodiments, the sample has been treated to amplify RNA prior to contact with the riboregulator. In some embodiments, the sample has been treated to amplify RNA isothermally prior to contact with the riboregulator.

Another related aspect of this disclosure provides a method of treating a subject, comprising administering an effective amount of an anti-viral agent to a subject having a viral infection, wherein the subject is identified as having a viral infection or as having been exposed to a virus by detecting viral mRNA in a sample from the subject using one or more of the foregoing toehold riboregulators, including for example a toehold riboregulator comprising a nucleotide sequence of any one of SEQ ID NOs: 43841, 9602, 62866, and 19367.

Another related aspect of this disclosure provides a method of treating a subject, comprising administering an effective amount of an anti-cancer agent to a subject having a cancer, wherein the subject is identified as having a cancer by detecting increased mRNA expression of a human transcription factor in a sample from the subject using a toehold riboregulator, including for example a toehold riboregulator comprising a nucleotide sequence of any one of SEQ ID NOs: 164989, 111698, and 236638.

Another related aspect of this disclosure provides a system for assisted design of RNA-based synthetic biology components comprising at least one pre-processing stage dedicated to transform input nucleic acid sequences into a multi-dimensional representation, at least one machine learning architecture trained and optimized for classification and/or regression of said pre-processed sequences to predict at least one experimentally measured performance metric, and at least one output representing the attention and/or saliency mechanisms exhibited by at least one of the said machine learning architectures to inform further design of RNA-based synthetic biology components.

Another related aspect of this disclosure provides a method for assisted design of RNA-based synthetic biology components comprising generating pre-processed sequences comprising transforming input nucleic acid sequences into a multi-dimensional representation, training and optimizing at least one machine learning architecture for classification and/or regression of said pre-processed sequences to predict at least one experimentally measured performance metric, and generating at least one output representing the attention and/or saliency mechanisms exhibited by at least one of the said machine learning architectures to inform further design of RNA-based synthetic biology components.

These and other aspects and embodiments will be described in greater detail herein.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1. Deep learning for RNA synthetic biology pipeline. RNA tool selection is followed by library synthesis and characterization with analysis using deep neural networks (DNN) to provide functionality predictions and design insight. We used a high-throughput toehold switch library as a canonical model for the general investigation of RNA synthetic biology tools. The original toehold switch architecture from Green et al. (2) was used, containing a 12-nucleotide (nt) toehold (a/a′) and an 18-nt stem (b/b′) fully unwound by the trigger (left-bottom). We selected to fuse the RNA trigger to the 5′ end of the switch by an unstructured linker to facilitate library synthesis. Then, a flow-sequence (seq) pipeline was used to characterize the fluorescence signal of individual toehold switches in a pooled sequential assay, including pooled induction, FACS sorting, next-generation sequencing (NGS) and count frequency analysis. Finally, various DNN architectures were used to predict data outputs, while features contributing to DNN predictions were intuitively visualized to elucidate biological insights.

FIGS. 2A-F. Flow-seq toehold switch library characterization and trigger ontology. The distribution of recovered toeholds for (A) ON-state signals, (B) OFF-state signals, and (C) calculated ON/OFF ratios are shown (selected from quality control process #3, QC3 in FIG. 18 and Table 1). (D) Validation results for toehold switches expressed in a PURExpress cell-free system with un-fused trigger RNA, including eight low-performing (poor, ON/OFF <0.05) and eight high-performing (good, ON/OFF >0.97) samples. Obtained flow-seq data show competency in performance-based switch classification within this distinct biological context. (E) Tested switch/trigger variants from each origin category, including randomly generated sequences, 906 human transcription factor transcripts, and 23 pathogenic viral genomes. (F) Experimental ON/OFF ratios for all triggers tiled across the transcripts of two clinically relevant human transcription factors (stat3 and kmt2a) upregulated in cancerous phenotypes (42, 43), as well as all triggers tiled across the genomes of two pathogenic viruses: West Nile Virus (WNV) and Human Immunodeficiency Virus (HIV). GFP=Green Fluorescent Protein; Seq=Sequence; HPV=Human Papillomavirus.

FIGS. 3A-G. Analysis of toehold switch performance using sequence k-mers, rational thermodynamic features, and sequence-based multilayer perceptron (MLP) models. (A) Sequence logos for k-mer motifs discovered to be disproportionately represented in weakly induced switches (low ON) and leaky switches (high OFF), functional proportions, and E-values. (B) The Pearson correlation (left, |max|=0.4) and R²metric (right, |max|=0.16) for thirty state-of-the-art thermodynamic features and obtained RBS Calculator v2.1 outputs. (C) Base architecture of investigated MLP models, featuring three fully connected layers. For training in regression-mode, three different outputs were predicted (ON, OFF, ON/OFF), whereas for classification training only a single binary output based on ON/OFF (threshold at 0.7) was predicted. (D) Box and whisker plots for R²between experimental and regression-based predictions for best performing rational features, logistic regression models and MLPs. (E) Minimum absolute error (MAE) between experimental and predicted values for these same models. (F) Box and whisker plots for area under the curve (AUC) of the receiver-operator curve (ROC) and the precision-recall curve (P-R) in classification-mode predictions compared to experimental values. In both regression and classification, the one-hot encoded sequence MLP delivered top-in-class performance without using pre-computed thermodynamic or kinetic metrics. (G) ROC curves of pre-trained MLP classification models validated with an unseen 168-sequence external dataset from Green et al. (2).

FIGS. 4A-D. Evaluation of neural network architectures with increased capacity. Performance metrics for convolutional neural networks (CNN) and long short-term memory (LSTM) networks trained on one-hot encoded toehold sequences, as well as a CNN trained on a two-dimensional, one-hot encoded sequence complementarity map. All models are compared to the previously reported MLPs trained on the 30 pre-calculated thermodynamic features and one-hot toehold sequences. For regression-based predictions (A) shows box and whisker plots for R²metric, while (B) shows MAE for all models. In the case of classification-based predictions (C) shows box and whisker plots of the area under the curve (AUC) of the receiver-operator curve (ROC) and the precision-recall curve (P-R) for all tested models. In both regression and classification, the one-hot encoded sequence MLP delivered top-in-class performance as compared to higher capacity deep learning models. (D) ROC curves of pre-trained higher-capacity classification models validated with an unseen 168-sequence external dataset from Green et al. (2). In A-C, for each group of 5 boxes, the first box represents MLP—All Rational Feat., the second box represents MLP—OneHot Seq., the third box represents CNN—OneHot Seq., the fourth box represents LSTM—OneHot Seq., and the fifth box represents CNN—2D Comp. Map.

FIGS. 5A-F. VIS4Map: Visualizing secondary structure features using saliency maps of a sequence-based complementarity matrix input. (A) A simplified schematic of the CNN-based architecture used to generate toehold functional predictions with network attention visualizations. The system receives a one-hot encoded, two-dimensional (2D) sequence complementarity map as input, followed by three 2D convolutional/max-pooling layers, a flattening step, and finally a set of dense layers. After output generation (e.g., OFF), a gradient-weighted activation mapping is performed to visualize activation maximization regions responsible for delivered predictions (VIS4Map). (B) Histograms of the percentage overlap between VIS4Maps generated from a CNN pre-trained to predict minimum free energy (MFE) using 120-nt RNA sequences and MFE maps generated by NUPACK. When analyzed using 500 random test set sequences, the distributions of correctly matched and randomly assigned maps are distinct with increased percentage overlap from matched samples as compared to unmatched. (C) Examples of saliency VIS4Maps compared with their corresponding MFE structures as predicted by NUPACK for three randomly selected 60-nt RNA sequences. See FIG. 16A for additional examples with 120-nt RNA sequences. (D) Four representative VIS4Map examples of randomly selected 118-nt RNA toehold switch sequences from an OFF-predictive CNN model. (E) Averaged VIS4Maps of 10,125 randomly selected toehold switch RNA sequences from our library test-set processed with our OFF-predicting CNN model (left) and compared their corresponding averaged MFE maps obtained using NUPACK (right). (F) Averaged VIS4Maps of the 10% most accurately predicted switches sorted by quartile from lowest OFF (tight) to highest OFF (leaky), inset at the hairpin stem, and toehold region of the switch. After contrast enhancement of averaged VIS4Maps to visualize sparsely distributed secondary structures, a noticeable increase in structures outside of the prominent equilibrium-designed switch hairpin structure appears, corresponding to increased toehold leakiness. A toehold switch schematic (right) is shown to denote how incorrectly folded and potentially weaker kinetically stable intermediate structures might compete with the correctly folded structure that is designed to be reached at equilibrium.

FIGS. 6A-B. Design and validation of oligomer library. Individual toehold switch constructs within the library were synthesized from a pool of oligomers, and a representative panel of constructs was verified against a previously published dataset. (A) Schematic of the pooled library oligo used for the synthesis of our high-throughput toehold switch library. Distinct toehold construct regions include: pre_seq (plasmid backbone sequence) (e.g., SEQ ID NO: 244020), promoter (T7 promoter including GGG) (e.g., SEQ ID NO: 244021), trigger (toehold-unique), switch (complete toehold and ascending stem), loop1 (region linking trigger to switch) (e.g., SEQ ID NO: 244022), loop2 (main toehold switch hairpin loop containing the RBS) (e.g., SEQ ID NO: 244023), stem1 (top half of descending stem), atg (start codon), stem2 (bottom half of descending stem), linker (21 nt sequence of unstructured amino acids) (e.g., SEQ ID NO: 244024) and post_linker (first 15 nt of GFP) (e.g., SEQ ID NO: 244025). Further detail can be found in Table 4. Amplification primers for both ON and OFF libraries (including the common reverse primer) are shown with black arrows. Sequences of common reverse primer, on forward primer and off forward primer are provided as SEQ ID NOs: 244026-244028. (B) Comparison of ON state GFP expression from a panel of 20 individually assayed switches from our high-throughput toehold switch pipeline against the ON/OFF ratio for equivalent switches reported by Green et al. (1). The agreement between the 5′ fused triggers used in this work and the separately transcribed triggers used by Green et al. (1) was assessed based on the Pearson correlation coefficient (0.8567). GFP=Green fluorescent protein, nt=nucleotide, RBS=Ribosome binding site.

FIG. 7. Library FACS distributions and empirically-derived sorting gates. To determine the boundaries of the sorting gates for our high-throughput toehold switch pipeline, we used Switch #4 from Green et al. (1) in ON and OFF conformations as positive controls, and a pUC19 plasmid lacking a GFP gene as a negative control. Fluorescence distribution plots of IPTG-induced E. coli BL21-star cells from the three control conditions are shown alongside complete ON and OFF libraries for comparison. Boundaries for the four sorting bins are shown as dotted lines.

FIGS. 8A-B. Inter-replicate variability of toehold switch libraries. For the same initial toehold library, we performed two replicates of the BL21 transformation process followed by independent induction, sorting, and sequencing. Two metrics were used to compare the inter-replicate variability: (A) the mean absolute error (MAE), and (B) the R²correlation coefficient. Shown are the MAE and R²values for ON (first bar in each pair) and OFF (second bar in each pair) measurements at different ranges of library count thresholds.

FIG. 9. Effect of QC level on MLP performance. The predictive power of our multilayer perceptron model was evaluated after training with datasets obtained from increasingly stringent quality control (QC) thresholds. The most stringent quality control group (QC5) was withheld as a test set, and an MLP trained on a one-hot representation of the toehold sequence was given either QC1 (first bar in each 4 bar group), QC2 (second bar in each 4 bar group), QC3 (third bar in each 4 bar group), or QC4 (fourth bar in each 4 bar group) as training data. From the resulting test-prediction of QC5 values, we show the MAE (upper panel), and the analogous R²correlation metric (lower panel) between the predicted and experimental values. See Table S1 for conditions for each QC level.

FIG. 10. Cell-free toehold switch validation. A panel of toeholds that showed either a low or high ON/OFF ratio as measured by our high-throughput flow-seq assay were individually cloned and assayed in a cell-free protein synthesis (CFPS) system. The time course velocities of GFP signal evolution are shown for the PURExpress CFPS reactions containing the sixteen switches with or without their separately transcribed RNA triggers. The sequences and flow-seq assay results for these sixteen switches can be found in Table 2. ON measurement is the first bar of each pair and corresponds to 10 μM trigger. OFF measurement is the second bar of each pair and corresponds to 0 μM trigger.

FIGS. 11A-B. Correlation between rational thermodynamic features and toehold switch dataset, subsetted for A-U content. We analyzed the R²coefficients between 30 commonly used thermodynamic features and the ON, OFF, or ON/OFF measurements of variants in our high-throughput dataset. (A) R²coefficients for the subset of switches that contained only an A-U or U-A base pair at the top of the toehold switch stem (positions 79 and 91 in Table 4). (B) R²coefficients for the entire set of switches, allowing for any base pair at the top of the toehold switch stem. Both R²value sets were compared to evaluate findings from Green et al. (1) where subsetting for switches with an A-U or U-A basepair at the top of the stem was sufficient to dramatically increase the predictive R²coefficient between thermodynamic features and measured ON/OFF. We found measurable differences between various thermodynamic features when subsetting for an A-U basepair at the top of the hairpin stem, particularly for those in the Ideal Ensemble Defect (ED) block. However, differences between the R²values in said subset and those obtained for other possible base-pairs were not statistically significant suggesting no overall increase in predictive value (p>0.05 for ON, OFF, and ON/OFF, two-tailed t-test).

FIGS. 12A-F. Kinetic toehold switch folding analysis using Kinfold. Folding trajectories were run using the Kinfold package for the OFF-state switch sequence (positions 50-134 nt in Table 4). (A) For a single representative toehold switch, six example trajectories are shown. Trajectories in green reached the MFE structure within 10³arbitrary time units (au), while those in blue did not. (B) For two representative toehold switches, 100 trajectories were run for a maximum time of 10⁶au. Histograms of the time required for a trajectory to reach the MFE structure are shown. Most trajectories took longer than 10³au, compared to the Kinfold analyses in Borujeni et al. (6), where average trajectory times fell in the range of 10¹-10³au, and 10⁴au was the longest allowed trajectory time. (C,D,E,F) For each switch in the QC4 dataset (total 19,983 variants), 100 trajectories were run and the following measurements plotted: (C) histograms of the mean and negative standard deviation of the trajectories' average energy during the first 10³au, (D) the fraction of trajectories that completed folding of the MFE structure before 10³au, (E) the ratio of average trajectory energy to the minimum possible MFE energy, and (F) the R²correlation between the metrics in C,D,E and the empirical measurements in our toehold switch dataset. For comparison with previous rational features the heatmap axis is set identically to FIG. 3B.

FIGS. 13A-C. Determination of the optimal ON/OFF binary classification cutoff threshold. AUC, P-R, and enrichment ratio analyses were used to determine the optimal cutoff threshold at which to binarize ON/OFF data for classification. We trained a standard MLP architecture on the one-hot sequence representation of the toehold switch at five different binarization thresholds, and compared the following performance metrics: (A) model AUROC results, (B) model AUPRC results, and (C) model enrichment ratio over random chance. The enrichment ratio is calculated as the fraction of true positive toehold switches returned by the model (i.e., the precision) divided by the fraction returned by random chance. The enrichment ratio was specifically calculated at the level of precision for which the recall returns one positive switch per 100, or approximately ten on average for a typical mRNA of length ˜1000 nt. The final threshold selected for all classification models in this study was 0.7 (or the top 8.3% of switches), balancing a high enrichment ratio with a practical degree of overall precision.

FIG. 14. MLP predictions vs. experimental results. Scatter plots of the predicted versus empirical values of our compiled test set are shown for ten-fold cross-validated MLP models trained with either the 30 pre-calculated rational thermodynamic features as inputs (left, dark green), or the toehold switch one-hot sequence representation as input (right, light green) for ON, OFF, and ON/OFF. Summary statistics are reported in FIG. 3D,E.

FIGS. 15A-D. Holdout validation of individual viral genomes. For each of the 23 pathogenic viruses tiled in our toehold switch dataset, every toehold switch targeting a given viral genome was withheld, and an MLP model was trained with the remaining sequences in the dataset using a one-hot sequence input representation classifying for ON/OFF ratio. The model performance was then evaluated on the switches of the withheld viral genoma as a test set. (A) Area under the receiver operating characteristic curves (AUROC) for holdout viral genomes. Dotted line denotes AUROC average across test samples. (B) Area under the precision-recall curves (AUPRC) for holdout viral genomes. Dotted line denotes AUPRC average across test samples. (C) Fraction of toehold switches in synthesized high-throughput library classified as high-performing for each virus type. Dotted line denotes average at 8%. (D) Total number of toehold switches synthesized for each virus type.

FIGS. 16A-B. VIS4Map analysis of random toehold sequences in MFE predictor 2D CNN model. A dataset of 50,000 random RNA sequences of length 120 nt and their corresponding MFE values were generated using NUPACK. A convolutional neural network (CNN) was then trained to predict the MFE of each sequence using either a one-hot representation or a complementarity map representation of the sequence as input. (A) For three randomly selected RNA sequences, representative saliency maps generated from the CNN model are shown alongside the MFE structure pre-computed independently using NUPACK. The CNN model was trained on complementarity map inputs. Overlap between salient diagonal features in the VIS4Map outputs and MFE structure maps is visible. (B) We then compared the R²coefficients between NUPACK-calculated MFE values and the predictions of a CNN model trained either on a one-hot representation or a complementarity matrix representation of the random RNA sequences. Error bars show standard deviation from five shuffled test sets.

FIG. 17. VIS4Map confusion matrix analysis of switch OFF conformation. Saliency maps generated from a CNN model trained to predict the toehold switch OFF metric are shown for different ground-truth OFF metrics. The model was trained using a complementarity matrix representation of the toehold sequence as input. Regions labeled on the axes are as follows: 1) Constant Loop, 2) Toehold, 3) Ascending Stem, 4) Constant RBS Loop, 5) Descending Stem, and 6) Constant Linker. Regions of interaction between constant regions are shaded darker as they do not contain variability between different switch sequences. All saliency maps were generated from the test set only. Saliency maps were then sorted according to the 25% highest and 25% lowest experimentally-determined OFF signal. The 10% best-predicted and 10% worst-predicted saliency maps from the high OFF and low OFF groups were then averaged to produce the shown confusion matrix. Contrast was enhanced four-fold in the averaged maps in order to visualize more sparsely distributed features.

FIG. 18. Dataset distribution vs. QC level. Histograms of toehold switch library values for ON, OFF, and ON/OFF were grouped according to our five different QC threshold levels and are shown here for comparison. The y-axis limits are held constant for ON, OFF, and ON/OFF, respectively, across QC levels after normalizing for data subset size.

Color version of these Figures are accessible on the United States Patent and Trademark Office PAIR website, under the Supplemental Tab of the file history for U.S. Provisional Application Ser. No. 62/948,175, filed Dec. 13, 2019.

DETAILED DESCRIPTION OF INVENTION

This disclosure provides numerous toehold riboregulators, each specific for a particular human transcription factor or a particular virus. Some of these riboregulators may be used to detect the presence of a particular virus, and this may aid in the diagnosis of an infection by such virus. Some of these riboregulators may be used to detect the presence or expression level of a particular human transcription factor, and this may aid in the diagnosis or prognosis of a condition associated with the presence and/or increased expression of such transcription factor. One such condition is cancer. For example, the human transcription factor STAT3 is reportedly upregulated in certain cancers, and it may therefore act as a diagnostic and/or prognostic marker of such cancers.

As will be described in greater detail herein, the toehold riboregulators may be provided covalently conjugated, typically at their 3′ ends, to a coding domain. The coding domain may be a reporter domain. The reporter domain may encode a reporter protein. Alternatively, the reporter domain may encode a reporter RNA (e.g., an RNA aptamer). Such toehold riboregulator-reporter domain constructs may be used to detect and/or measure a level (e.g., an expression level) of a nucleic acid of interest (i.e., a trigger nucleic acid that is present in the sample being tested).

Riboregulators are nucleic acid molecules that exist in two different conformations (i.e., closed and open conformations). In the closed conformation, the riboregulator adopts a secondary hairpin structure that sequesters a ribosome binding site (RBS) in a loop domain, rendering the RBS inaccessible to translation machinery. In the open conformation, the riboregulator adopts a linear structure and the RBS is no longer sequestered and rather it is accessible to the translation machinery. Riboregulators are designed to convert from their closed to their open conformations in the presence of a target nucleic acid (referred to herein as a trigger nucleic acid), which is typically the nucleic acid of interest in a sample. Thus, the conversion from closed to open conformations occurs upon specific binding of the riboregulator to a trigger nucleic acid. The binding of to the trigger causes the conversion which then enables expression of a downstream coding domain, such as a reporter protein domain. Presence of the reporter protein is therefore a surrogate for the presence of the trigger nucleic acid.

The riboregulators share a common structure, as shown in FIG. 1, which includes, in a 5′ to 3′ order, a single-stranded toehold domain, a hairpin domain comprising a stem domain and a loop domain, and a linker domain. The loop domain comprises the ribosome binding site (RBS, e.g., AGAGGAGA) and the stem domain comprises the start codon, AUG, on the descending strand. The stem domain further comprises first and second stem domains, called stem domain 1 and stem domain 2. The start codon, AUG, separates stem domain 1 from stem domain 2, on the descending strand, as illustrated in FIG. 1. The single-stranded toehold domain and the adjacent ascending sequence of the stem domain together are referred to as the “switch” domain. This domain is represented by a+b (5′ to 3′) sequences in FIG. 1.

It is this switch domain which is complementary to the “trigger” nucleic acid being detected, which as described above is either a particular viral nucleic acid or a nucleic acid encoding a particular human transcription factor acid. The trigger is represented by a′+b′ (3′ to 5′) sequences in FIG. 1. The trigger sequence a′ hybridizes by complementary base pairing to the single-stranded toehold domain denoted a. As the stem domain naturally associates and dissociates (i.e., “breathes”), trigger sequence b′ then has the opportunity to migrate into the hairpin and hybridize to the ascending sequence of the stem domain, denoted b. This hybridization unwinds the stem domain further and makes the RBS accessible to the ribosome, and this in turn enables translation of the downstream coding region, and production of the encoded protein (e.g., the reporter protein). In the absence of a trigger nucleic acid, the toehold domain remains single-stranded, the riboregulator typically remains closed, the RBS remains inaccessible to the ribosome machinery, and there is no or little translation of the downstream coding domain.

As will be understood, in their final form, riboregulators are RNA molecules that possess an RBS and are acted upon by ribosome machinery to produce an encoded protein. While they may be provided to a system, such as a cell-free system or an in vivo system, as RNAs, this is likely to be inefficient given the inherent instability of RNA. Instead, they are typically provided in a DNA form, conjugated to a promoter, such as but not limited to a T7 promoter, and are then produced in an RNA form through transcription from the T7 promoter. The sequences provided in the sequence listing submitted herewith and as part of this specification are DNA sequences that comprise the riboregulator sequence in a DNA form (i.e., there is a T in the sequence provided whereas the RNA counterpart would have a U in that position). Thus, these sequences are understood to comprise the DNA form of a riboregulator (with Ts) as well as the RNA form (with Us). As will be discussed below, these sequences also comprise elements in addition to the riboregulator elements discussed above.

The nucleic acids provided as SEQ ID NOs: 1-244,000 are DNAs that comprise the riboregulator elements described above. These sequences have a common structure/sequence as follows, in a 5′ to 3′ order:

Promoter sequence: T7 promoter, 20 nt,

(SEQ ID NO: 244001)

TAATACGACTCACTATAGGG;

- Switch domain sequence: complete toehold (12 nt) and entire ascending stem (18 nt), 30 nt in total; variable sequence;

Loop domain sequence: includes RBS, 11 nt,

(SEQ ID NO: 244002)

AACAGAGGAGA;

- Stem domain 1 sequence: top half of descending stem, 6 nt, variable sequence will be dictated by switch domain sequence, as illustrated in FIG. 6A;
- ATG or AUG: start codon, 3 nt;
- Stem domain 2 sequence: bottom half of descending stem, 9 nt, variable sequence will be dictated by switch domain sequence, as illustrated in FIG. 6A;
- Linker domain sequence: sequence encoding unstructured amino acids, 21 nt,

(SEQ ID NO: 244003)

AACCTGGCGGCAGCGCAAAAG;

- Post-linker sequence: ATG start of reporter gene, 3 nt.

Accordingly, each of the sequences in the enclosed sequence listing is 103 nucleotides in length. These nucleic acid sequences are provided as DNA strands, which are then transcribed from the T7 promoter into RNA strands which are able to self-hybridize and thereby adopt the riboregulator structure described above. Further, before use, each of these sequences may be conjugated (i.e., operably linked) to a coding domain at their 3′ ends. These sequences may be provided in a replication vector and/or an expression vector, and optionally in a host cell.

This disclosure contemplates use of the entire 103 nt sequence, for example by conjugating such sequence to a coding domain. Alternatively, this disclosure contemplates use of the sequence presented by nucleotides 21-100, which represent the toehold domain, the hairpin domain, and the linker domain, preferably in RNA form (i.e., with Ts replaced with Us and with an RNA backbone).

A cell or a cell-free system may be contacted with the riboregulator in its DNA form, and it may be transcribed from the T7 promoter in order to form its RNA form. A sample to be tested may be contacted with the DNA form, provided such sample is capable of transcribing the DNA form. Alternatively, the sample to be tested may be contacted with the RNA form, and thus the sample to be tested may be contacted with a riboregulator RNA sequence beginning at the switch domain and having a coding domain. In relation to SEQ ID NOs: 1-244,000, this means that samples may be contacted with RNA versions of these sequences that lack nucleotides 1-20 but that comprise a coding domain conjugated to their 3′ ends.

Use of riboregulators in vitro as nucleic acid diagnostic tools using cell-free protein synthesis (CFPS) systems have been described previously (3, 4, 15, 16), and reference can be made to such prior teachings.

It is to be understood therefore that in its RNA form, the riboregulator typically lacks the promoter sequence and it is conjugated to a coding domain as shown in FIG. 1.

It is also to be understood that other promoters may be used in place of the T7 promoter that is provided in SEQ ID NOs: 1-244,000.

The consensus sequence therefore comprises certain constant or invariant sequences including the promoter sequence, the loop domain sequence, the linker sequence, and the post-linker sequence. The switch domain sequence, the stem domain 1 sequence, and the stem domain 2 sequence are all variable (i.e., they will vary between riboregulators), although they will have regions of complementarity to each other. This is illustrated in FIG. 6A which shows that stem domain 1 sequence is complementary to a 3′ region of the switch domain, and stem domain 2 sequence is complementary to a 5′ region of the switch domain. The complementarity of these regions results in the stem domain.

In general, the hairpin and stem domains described herein form at and are stable under physiological conditions, e.g., conditions present within a cell (e.g., conditions such as pH, temperature, and salt concentration that approximate physiological conditions). Such conditions include a pH between 6.8 and 7.6, more preferably approximately 7.4. Typical temperatures are approximately 37° C.

Various of the nucleic acids provided in this disclosure may be regarded as non-naturally occurring, artificial, engineered or synthetic. This means that the nucleic acid is not found naturally or in naturally occurring, unmanipulated, sources. A non-naturally occurring, artificial, engineered or synthetic nucleic acid may be similar in sequence to a naturally occurring nucleic acid but may contain at least one artificially created insertion, deletion, inversion, or substitution relative to the sequence found in its naturally occurring counterpart. A cell that contains an engineered nucleic acid may be regarded as an engineered cell.

In some instances, the riboregulators are operably linked to coding regions that encode reporter proteins. Such reporter proteins are typically used to visualize activation of the riboregulator and thus presence of the trigger nucleic acid in the sample being analyzed. Reporter proteins suitable for this purpose include but are not limited to fluorescent or chemiluminescent reporters (e.g., GFP variants, luciferase, e.g., luciferase derived from the firefly (Photinus pyralis) or the sea pansy (Renilla reniformis) and mutants thereof), enzymatic reporters (e.g., β-galactosidase, alkaline phosphatase, DHFR, CAT), etc. The eGFPs are a class of proteins that has various substitutions (e.g., Thr, Ala, Gly) of the serine at position 65 (Ser65). The blue fluorescent proteins (BFP) have a mutation at position 66 (Tyr to His mutation) which alters emission and excitation properties. This Y66H mutation in BFP causes the spectra to be blue-shifted compared to the wtGFP. Cyan fluorescent proteins (CFP) have a Y66W mutation with excitation and emission spectra wavelengths between those of BFP and eGFP. Sapphire is a mutant with the suppressed excitation peak at 495 nM but still retaining an excitation peak at 395 and the emission peak at 511 nM. Yellow FP (YFP) mutants have an aromatic amino acid (e.g. Phe, Tyr, etc.) at position 203 and have red-shifted emission and excitation spectra.

The riboregulators comprise an RBS. Exemplary RBS sequences include, but are not limited to, AGAGGAGA (or subsequences of this sequence, e.g., subsequences at least 6 nucleotides in length, such as AGGAGG). Shorter sequences are also acceptable, e.g., AGGA, AGGGAG, GAGGAG, etc. Numerous synthetic ribosome binding sites have been created, and their translation initiation activity has been tested. The activity of any candidate sequence to function as an RBS may be tested using any suitable method. For example, expression may be measured as described in Example 1 of published PCT application WO 2004/046321, or as described in reference 53 of that published PCT application, e.g., by measuring the activity of a reporter protein encoded by an mRNA that contains the candidate RBS appropriately positioned upstream of the AUG.

Particular Riboregulators

Some of the riboregulators of this disclosure are specific for (i.e., they specifically hybridize to, and thus can be used to detect) nucleic acids (DNA or RNA) from particular viruses. These viruses are astrovirus, cardiovirus, chikungunya virus, cosavirus, coxsackie virus, dengue virus, ebola virus, hantavirus, human immunodeficiency virus, human parvo virus, human rhino virus, influenza virus: h1n1, influenza virus: h3n2, lassa virus, leishmanial virus, Marburg virus, papilloma virus, poliovirus, rabies virus, smallpox virus, west nile virus, yellow fever virus, and zika virus. The switch domain of these virus-specific riboregulators will hybridize to a nucleic acid, such as a transcript, from one of these viruses. Table 5 provides details relating to the SEQ ID NO: viral specificity.

Of particular interest are riboregulators having a nucleotide sequence selected from the group consisting of SEQ ID NOs: 43841, 9602, 62866, and 19367. Of particular interest are riboregulators that are specific for smallpox virus, dengue virus, and human rhino virus.

Some of the riboregulators of this disclosure are specific for (i.e., they specifically hybridize to, and thus can be used to detect and optionally measure) nucleic acids (DNA or RNA) that encode particular human transcription factors. These human transcription factors are AC097634.4, ACTB, ACTL6A, ACTN4, AEBP1, AEBP2, AGO1, AGO2, AHR, AIRE, AKNA, AL121581.1, ALX1, ALX4, ANHX, AR, ARHGAP35, ARID3A, ARID3B, ARID3C, ARID4A, ARID4B, ARID5A, ARID5B, ARNT, ARNT2, ARNTL, ARNTL2, ARRB1, ARX, ASCL1, ASCL2, ASCL3, ASCL4, ASCL5, ASH2L, ATF1, ATF2, ATF3, ATF4, ATF5, ATF6, ATF6B, ATMIN, ATOH1, ATOH8, ATXN3, BACH1, BACH2, BARHL1, BARHL2, BARX1, BARX2, BASP1, BATF, BATF2, BATF3, BAZ2A, BCL11A, BCL11B, BCL6, BCL6B, BCOR, BHLHA15, BHLHE40, BHLHE41, BORCS8-MEF2B, BRCA1, BRD7, BRF2, CALCOCO1, CARF, CARM1, CBX4, CC2D1A, CC2D1B, CCAR1, CCNT1, CDC5L, CDK12, CDK13, CDK5RAP2, CDK9, CDX1, CDX2, CDX4, CEBPA, CEBPB, CEBPD, CEBPE, CEBPG, CEBPZ, CGGBP1, CHD2, CHD4, CHD7, CIART, CIITA, CITED1, CLOCK, CNBP, CREB1, CREB3, CREB3L1, CREB3L2, CREB3L3, CREB3L4, CREBBP, CREBRF, CREM, CRX, CRY1, CRY2, CT476828.9, CTCF, CTCFL, CUX1, CUX2, CXXC1, DACH1, DBP, DDIT3, DDN, DEAF1, DHX36, DHX9, DLX1, DLX2, DLX4, DLX5, DMBX1, DMRT1, DMRT2, DNMT3A, DPF2, DR1, DRAP1, DUX4, E2F1, E2F2, E2F3, E2F4, E2F6, E2F7, E2F8, E4F1, EAF2, EBF2, EBF3, EBF4, EED, EGR1, EGR2, EGR3, EGR4, EHF, EHMT2, ELF1, ELF3, ELF4, ELF5, ELK1, ELK3, ELK4, ELL3, ELMSAN1, EN1, ENO1, EOMES, EP300, ERBB4, ERG, ESR1, ESR2, ESRRA, ESRRB, ESRRG, ESX1, ETS1, ETS2, ETV1, ETV2, ETV3, ETV4, ETV5, ETV6, ETV7, EZH2, FERD3L, FEZF1, FEZF2, FIGLA, FLI1, FOS, FOSB, FOSL1, FOSL2, FOXA1, FOXA2, FOXA3, FOXC1, FOXC2, FOXD1, FOXD3, FOXF1, FOXF2, FOXH1, FOXI1, FOXJ1, FOXJ2, FOXK1, FOXK2, FOXL2, FOXM1, FOXN4, FOXO3, FOXP2, FOXP3, FOXQ1, FOXS1, FUBP3, GABPA, GABPB1, GABPB2, GADD45A, GATA1, GATA2, GATA3, GATA4, GATA5, GATA6, GATAD2B, GBX2, GCFC2, GCM1, GFI1, GLI1, GLI2, GLI3, GLIS1, GLIS2, GLMP, GMEB1, GMEB2, GRHL1, GRHL2, GSC, GSX1, GTF2B, GTF3C1, GZF1, H2AFY, H2AFY2, H2AFZ, H3F3A, H3F3B, HAND1, HAND2, HDAC1, HDAC2, HDAC4, HDAC5, HDAC6, HELT, HES1, HES2, HES3, HES4, HES5, HES6, HES7, HESX1, HEY1, HEY2, HEYL, HHEX, HIC2, HIF1A, HINFP, HIVEP1, HLF, HLTF, HMGA1, HMGA2, HMGB1, HMGB2, HMX1, HMX3, HNF1A, HNF1B, HNF4A, HNF4G, HNRNPC, HNRNPK, HNRNPL, HNRNPU, HOXA10, HOXA2, HOXA3, HOXA4, HOXA5, HOXA6, HOXA7, HOXA9, HOXB1, HOXB2, HOXB3, HOXB4, HOXB5, HOXB6, HOXB7, HOXB9, HOXC10, HOXC11, HOXC4, HOXC5, HOXC6, HOXD10, HOXD13, HOXD3, HOXD4, HOXD8, HOXD9, HR, HSF1, HSF2, HSF4, HSF5, HSFX1, HSFX2, HSFX3, HSFX4, HSFY1, HSFY2, IER2, IFI16, IKZF1, IKZF2, IKZF3, IKZF4, IKZF5, INSM1, IRF1, IRF2, IRF2BP1, IRF2BP2, IRF2BPL, IRF3, IRF4, IRF5, IRF6, IRF7, IRF8, IRF9, ISL1, JARID2, JDP2, JMJD1C, JUN, JUNB, JUND, KAT2B, KAT7, KCNIP3, KDM1A, KDM2B, KDM3A, KDM3B, KDM5A, KDM6A, KDM6B, KLF1, KLF10, KLF11, KLF12, KLF13, KLF15, KLF16, KLF17, KLF3, KLF4, KLF5, KLF6, KLF7, KLF8, KMT2A, KMT2D, LDB1, LEF1, LHX2, LHX3, LITAF, LMO2, LMO4, LMX1A, LMX1B, LONP1, LRRFIP1, LYL1, MACC1, MAF, MAF1, MAFA, MAFB, MAFF, MAFG, MAFK, MAX, MAZ, MBD2, MBD3, MED1, MED12, MED8, MEF2A, MEF2B, MEF2C, MEF2D, MEIS1, MEIS2, MEN1, MEOX1, MEOX2, MESP1, MESP2, MITF, MIXL1, MLX, MLXIP, MLXIPL, MMP12, MNT, MRTFA, MSC, MSGN1, MSX1, MSX2, MTA1, MTA2, MTERF3, MTF1, MTF2, MTOR, MUC1, MXD1, MXD3, MXI1, MYB, MYBBP1A, MYBL1, MYBL2, MYC, MYCN, MYEF2, MYF5, MYF6, MYOCD, MYOD1, MYOG, MYPOP, MYT1, MYT1L, MZF1, NACC2, NANOG, NCOA2, NCOR1, NCOR2, NDN, NEUROD1, NEUROD2, NEUROD6, NEUROG1, NEUROG2, NEUROG3, NFAT5, NFATC1, NFATC2, NFATC3, NFATC4, NFE2, NFE2L1, NFE2L2, NFE2L3, NFIA, NFIB, NFIC, NFIL3, NFKB1, NFKB2, NFX1, NFXL1, NFYA, NFYB, NFYC, NHLH1, NHLH2, NKRF, NKX2-1, NKX2-2, NKX2-5, NKX2-6, NKX2-8, NKX3-1, NKX3-2, NKX6-1, NKX6-2, NLRC5, NME1, NONO, NOTCH1, NPAS2, NPAS4, NPM1, NR1D1, NR1D2, NR1H2, NR1H3, NR1H4, NR1I2, NR1I3, NR2C1, NR2C2, NR2E3, NR2F1, NR2F6, NR3C1, NR4A1, NR4A2, NR4A3, NR5A1, NR5A2, NR6A1, NRF1, NRIP1, NRL, NSD1, ONECUT2, ONECUT3, OSR1, OSR2, OTX1, OTX2, OVOL1, PARP1, PATZ1, PAX1, PAX2, PAX4, PAX5, PAX6, PAX8, PAX9, PAXBP1, PBX1, PBX2, PBX3, PCGF3, PCGF5, PCGF6, PDX1, PER1, PER2, PER3, PGR, PHB, PHOX2A, PHOX2B, PIH1D1, PITX1, PITX2, PITX3, PKNOX2, PLAG1, PLAGL1, POLRMT, POU1F1, POU2AF1, POU2F1, POU2F2, POU2F3, POU3F2, POU3F4, POU4F1, POU4F2, POU4F3, POU5F1, POU6F1, PPARA, PPARD, PPARG, PRDM1, PRDM11, PRDM12, PRDM13, PRDM14, PRDM15, PRDM2, PRDM4, PRDM5, PRDM6, PRDM7, PRDM9, PRDX5, PRKN, PRMT5, PROP1, PROX1, PRRX1, PSPC1, PTF1A, PURA, PURB, PURG, RAI1, RARA, RARB, RARG, RAX, RAX2, RB1, RBBP4, RBBP5, RBL1, RBL2, RBMX, RBPJ, RBPJL, RCOR1, RCOR2, RCOR3, REL, RELA, RELB, REST, RFX1, RFX2, RFX3, RFX4, RFX5, RFX6, RFX7, RFX8, RNF10, RORA, RORB, RORC, RPS3, RPTOR, RREB1, RRN3, RUNX1, RUNX2, RUNX3, RUVBL2, RXRA, RXRB, SAFB, SALL1, SALL2, SARS, SATB1, SATB2, SCRT1, SCRT2, SCX, SETX, SFPQ, SIN3A, SIRT1, SIX1, SIX2, SIX3, SIX4, SIX5, SIX6, SKIL, SMAD1, SMAD2, SMAD3, SMAD4, SMAD5, SMAD6, SMAD7, SMARCA2, SMARCA4, SMARCB1, SMARCC1, SMARCC2, SMARCD2, SMARCE1, SMYD3, SNAI1, SNAI2, SNAI3, SNCA, SOX1, SOX10, SOX11, SOX12, SOX13, SOX17, SOX18, SOX2, SOX21, SOX3, SOX4, SOX6, SOX7, SOX8, SOX9, SP1, SP2, SP3, SP5, SP7, SPI1, SPIB, SPIC, SREBF1, SREBF2, SRF, SSBP2, SSBP3, SSBP4, ST18, STAT1, STAT3, STAT5B, STAT6, STOX1, SUV39H1, SUV39H2, SUZ12, TAF1, TAF1B, TAF1C, TAF2, TAF5, TAF7, TAF7L, TAF9, TAF9B, TAL1, TAL2, TBL1X, TBL1XR1, TBP, TBPL1, TBPL2, TBR1, TBX15, TBX18, TBX19, TBX2, TBX20, TBX21, TBX22, TBX3, TBX5, TBX6, TBXT, TCF12, TCF15, TCF20, TCF21, TCF3, TCF4, TCF7, TCF7L1, TCF7L2, TCFL5, TEAD1, TEAD2, TEAD3, TEAD4, TEF, TFAM, TFAP2A, TFAP2B, TFAP2C, TFAP2D, TFAP2E, TFAP4, TFCP2, TFCP2L1, TFDP1, TFDP2, TFE3, TFEB, TFEC, TGIF1, THAP1, THAP11, THRA, THRAP3, THRB, TIPARP, TLX1, TNF, TOP1, TOX2, TOX3, TP53, TP63, TP73, TRERF1, TRIM24, TRPS1, TWIST1, TXK, UBTF, UHRF1, USP3, UTY, VAX1, VAX2, VDR, VEZF1, WBP2, WNT1, WNT11, WNT5A, WT1, XBP1, XRCC5, XRCC6, XRN2, YAP1, YBX1, YBX3, YY1, YY2, ZBED1, ZBTB14, ZBTB16, ZBTB17, ZBTB2, ZBTB20, ZBTB24, ZBTB4, ZBTB48, ZBTB5, ZBTB7A, ZBTB7B, ZC3H4, ZC3H6, ZC3H8, ZEB1, ZFHX2, ZFHX3, ZFHX4, ZFP42, ZFPM1, ZGPAT, ZHX3, ZIC1, ZIC2, ZIC3, ZIC4, ZIC5, ZKSCAN3, ZNF131, ZNF143, ZNF148, ZNF174, ZNF175, ZNF202, ZNF205, ZNF217, ZNF219, ZNF239, ZNF277, ZNF281, ZNF322, ZNF335, ZNF350, ZNF395, ZNF431, ZNF497, ZNF501, ZNF513, ZNF516, ZNF536, ZNF541, ZNF564, ZNF568, ZNF589, ZNF605, ZNF613, ZNF639, ZNF649, ZNF658, ZNF668, ZNF691, ZNF692, ZNF704, ZNF709, ZNF711, ZNF740, ZNF746, ZNF750, ZNF821, ZNF835, ZNF93, and ZSCAN21. Table 5 provides details relating to the SEQ ID NO: transcription factor specificity.

In some embodiments, the riboregulator is specific for STAT3 transcription factor, and it is used to detect and optionally measure the expression level of this transcription factor. Riboregulators specific for STAT3 are provided as SEQ ID NOs: 210632-210860.

Of particular interest are riboregulators having a nucleotide sequence selected from the group consisting of SEQ ID NOs: 164989, 111698, and 236638. Of particular interest are riboregulators that are specific for transcription factors NCOR1, E2F3 and ZNF175.

This disclosure refers to riboregulators that are specific for a particular virus or a particular transcription factor. This intends that the switch domain in such riboregulators is complementary to a nucleic acid sequence in or produced from the particular virus or to a nucleic acid coding for the particular transcription factor. The nucleic acid sequence in or produced from the particular virus or the nucleic acid coding for the particular transcription factor are considered triggers in this disclosure.

This disclosure contemplates variants of the riboregulators provided herein. For example, the disclosure contemplates variants that differ from the disclosed sequences by 1, 2, 3, 4 or 5 nucleotides, wherein such variants retain the ability to specifically hybridize to the original trigger of interest (i.e., the trigger of their parent riboregulator). Such variants may have a cell-free ON/OFF value that less than that of their parent riboregulator provided that such ON/OFF value is still suitable for use. The ON/OFF value may be for example 2, 3, 4, 5, or more.

The riboregulators may be defined by their strength, and this in turn may be defined by the level of expression of the coding domain in the presence (ON state) versus in the absence (OFF state) of the trigger nucleic acid. The riboregulators may have a cell-free ON/OFF value of about 2 to about 10, and may be further subdivided into those having an ON/OFF value of about 2 to about 5 and about 5 to about 8 and about 8 to about 10. In some instances, riboregulators with higher ON/OFF may be preferred. The ON/OFF of an individual riboregulator in a cell-free system may be determined as described in the Examples.

Applications/Uses

The riboregulators may be used in a number of applications. For example, they may be used to detect presence of nucleic acid such as an RNA in a sample, and such a method may comprise combining any one or a combination (e.g., 2) of the toehold riboregulators provided herein with a sample, wherein the riboregulator comprises a switch domain including a single-stranded toehold domain that is complementary to a nucleic acid (e.g., an RNA) in the sample, such as a nucleic acid encoding a transcription factor a viral protein (e.g., a transcription factor RNA or a viral RNA). The riboregulator comprises a coding domain that encodes a reporter protein, under conditions that allow translation of the coding domain in the presence of the nucleic acid (e.g., RNA) of interest but not in the absence of such nucleic acid (e.g., RNA). The method further comprises detecting the reporter protein as an indicator (or surrogate) of the nucleic acid (e.g., RNA) of interest. As used herein, conditions that allow translation of the coding domain are conditions that include all the necessary machinery to produce a protein from an RNA such as but not limited to ribosomes, tRNAs, and the like.

Samples to be tested include samples obtained from a subject. The subject may be a human or a non-human.

In some instances, the subject is a subject having, suspected of having, or at risk of having a condition associated with the presence of a particular viral nucleic acid (e.g., a viral RNA) such as an infection by one of the viruses listed above. Thus, for example, the subject may be a subject having, suspected of having, or at risk of having an astrovirus infection, a cardiovirus infection, a chikungunya virus infection, a cosavirus infection, a coxsackie virus infection, a dengue virus infection, an ebola virus infection, a hantavirus infection, a human immunodeficiency virus infection, a human parvo virus infection, a human rhino virus infection, an influenza h1n1 virus infection, an influenza h3n2 virus infection, a lassa virus infection, a leishmanial virus infection, a Marburg virus infection, a papilloma virus infection, a polio virus infection, a rabies virus infection, a smallpox virus infection, a west nile virus infection, a yellow fever virus infection, or a zika virus infection.

In some instances, the subject is a subject having, suspected of having, or at risk of having a condition associated with the presence and optionally increased expression of a particular human transcription factor from the list provided herein. A condition associated with the presence and optionally increased expression of a particular human transcription factor from the list provided herein is cancer.

In some embodiments, the transcription factor is STAT3 and the cancer is epithelial cancer such as squamous cell carcinoma of the head and neck, breast, ovary, prostate or lung cancer. In some embodiments, the cancer is intrahepatic cholangiocarcinoma. The presence and/or expression level of STAT3 may be used to diagnose or to prognose a particular cancer.

In some instances, the disclosure contemplates use of more than one virus-specific riboregulator. For example, some methods may involve contacting a sample with a plurality of virus-specific riboregulators in order to detect the presence of a plurality of viruses at the same time, or at least to test for the presence of a plurality of viruses at the same time. In this way, a single sample may be used and screened for the presence of a number of viruses. In order to distinguish which virus(es) are present in the sample, the riboregulators may be distinguished from each other based on the reporter protein to which they are operably linked. For example, GFP may be used as the reporter protein for HIV specific riboregulators.

The disclosure further contemplates that one or more riboregulators specific for the same virus may be used together. This may help with increasing the sensitivity of the detection assay. For example, riboregulators having SEQ ID Nos: 43841 and 62866 may be used together to detect smallpox virus. In some instances, the riboregulators are physically separate and drive translation of their respective reporter protein. In other instances, the riboregulators are physically attached, for example as an AND or an OR gate, and may contributed collectively to translation of a single reporter protein. Reference can be made to published PCT application WO 2014/074648 for a discussion of AND OR gates in the context of concatenated riboregulators.

In some embodiments, the riboregulator is operably linked to a coding domain that encodes a suicide gene (or suicide protein). In this way, the riboregulator can be used to selectively kill cells that are infected with a particular virus selected from the list provided herein. Alternatively, the riboregulator can be used to selectively kill cells that have increased expression of a particular transcription factor, such as STAT3, and which may therefore be cancer cells or pre-cancerous cells. An exemplary suicide gene is thymidylate synthase, and a subject is administered ganciclovir following production of the thymidylate synthase. In some embodiments, the suicide gene is herpes simplex virus type 1 thymidine kinase (HSV1-TK).

The riboregulators may be used to detect targets of interest such as viruses, and thus diagnose exposure to or infection by such viruses. The riboregulators may be used with an unmanipulated sample. Alternatively, the sample may be processed prior to contact with the riboregulator. For example, the sample may processed in order to extract RNA. Additionally or alternatively, the sample may be process to amplify RNA.

There are various techniques, including isothermal techniques, for amplifying nucleic acids such as RNA. One such method, referred to as nucleic acid sequence based amplification (NASBA)-mediated RNA amplification, is described by Pardee et al. Cell, 165:1255-1266, 2016. For example, RNA may be amplified using a method that comprises reverse transcription of a target RNA of interest using a sequence-specific reverse primer to form an RNA/DNA duplex. This duplex is then contacted with RNase H to degrade the RNA template. A forward primer having a T7 promoter is then introduced and allowed to bind and initiate elongation from the complementary strand, to form a double-stranded DNA product. T7-mediated transcription is then used to generate copies of the target RNA. NASBA is initiated at a higher temperature (e.g., about 65° C.) and then followed by isothermal amplification at about 41° C.

When used together, the isothermal RNA amplification and riboregulator-mediated detection steps provide a relatively low-cost and low-resource detection strategy.

The step of contacting the sample with the riboregulator can be performed in solution. Alternatively it can be performed in a paper-based form, as described by Pardee et al. Cell, 165:1255-1266, 2016.

EXAMPLES

In order that the invention described herein may be more fully understood, the following examples are set forth. It should be understood that these examples are for illustrative purposes only and are not to be construed as limiting this invention in any manner.

Engineered RNA modules are programmable elements capable of detecting small molecules, proteins, and nucleic acids. While useful, predicting the behavior of these tools remains a challenge, a situation that could be addressed through enhanced pattern recognition from deep learning. Thus, we investigate Deep Neural Networks (DNN) to predict toehold switch function as a canonical riboswitch model in synthetic biology. To facilitate DNN training, we synthesized and characterized in vivo a dataset of 91,534 toehold switches spanning 23 viral genomes and 906 human transcription factors. DNNs trained on nucleotide sequences (R²=0.43) outperformed previous state-of-the-art thermodynamic and kinetic models (R²=0.0001-0.04) and allowed for human-understandable attention-visualizations (VIS4Map) to identify failure modes. This deep learning pipeline constitutes a major step forward in engineering and understanding of RNA synthetic biology.

Materials and Methods

Toehold Switch Architecture Selection

The first-generation toehold switch architecture from Green et al. (SI-1) was selected in order to maximize the sequence variability in switch regions contributing to secondary structure. Where in later designs the trigger RNA only unwound a fraction of the stem (SI-1-3), in this earlier design the entire hairpin stem was variably complementary to the trigger, increasing the diversity of characterized RNA hairpins (FIG. 1A). An alternative fused ON state was also utilized. Normally, toehold switches detect the presence of a separate trigger RNA transcribed in trans to the OFF-state switch mRNA. However, for the testing of a large library of toehold-switch pairings, a two-plasmid system becomes intractable because each switch is designed around a specific cognate trigger. A two-plasmid system can also increase stochasticity caused by copy number variability. Green et al. (SI-1) found a strong positive correlation between conditions when the trigger is fused to the switch and conditions when un-fused, separate triggers are transcribed in excess. We confirmed this correlation ourselves on a subset of twenty toehold switches by comparing the signal from the alternative fused ON state used in our library to the measured ON/OFF from Green et al. (SI-1). Green et al have stated that due to a low switch plasmid copy number their OFF state rarely exceeded background autofluorescence, meaning that their reported ON/OFF ratios are essentially ON state measurements. The resulting comparison of signal from the alternative fused ON state we measured and the un-fused ON state measured by Green et al using a two-plasmid system resulted in a Pearson R=0.8567, as seen in FIG. 6B. Thus, the ON state of the switch can be reliably approximated by fusing the trigger RNA to the 5′ end of the switch mRNA using a constant, unstructured linker sequence (FIG. 1A, 6A), allowing for the direct synthesis of trigger-switch cognates on a single plasmid.

Library Trigger Sequence Selection

Viral genomes were obtained on Nov. 6, 2018, from the NCBI/NIH website (genome/viruses). Each retrieved genome was tiled 30 bp at a time (the trigger length), with a stride of 5 bp, spanning the respective genome. Human transcription factors were obtained using ENSEMBL 94 BioMart (SI-4) utilizing the Gene Ontology term GO:0044212 (transcription regulatory region DNA binding). The coding region of each transcription factor was tiled 30 bp at a time with a stride of 10. A remaining portion of the designs (˜10,000) was based on random 30 bp triggers.

Toehold Library Synthesis

We designed 244,000 toehold switch variants using 230 bp oligos, which were ordered and synthesized by Agilent. For each toehold switch variant, the oligo was designed containing the following sequence components in order from 5′ to 3′: 20 nt of common backbone, a T7 Promoter, the 30 nt Trigger sequence, a 20 nt unstructured Linker, the 12 nt Toehold, the 18 nt Ascending Stem, a 11 nt SD-containing Loop, the 18 nt Descending Stem including the start codon, a 21 nt AA-Linker, and the first 15 nt of the GFP gene. A schematic of the design can be found in FIG. 6A. In the previous validation of the fused trigger approach by Green et al. (SI-1), only part of the trigger was fused to avoid recombination of long repeated sequences, but the nature of our flow-seq pipeline allowed us to avoid this issue since the integrity of all variants was confirmed after measuring fluorescence through next-generation sequencing (NGS). The oligos were received at a stock amount of 10 pmol, which we diluted in 500 uL TE buffer for a working concentration of 20 nM. Of this working stock, 0.25 uL was used in 50 uL qPCR reactions using NEB Q5 polymerase 2×MM with 50 nM final concentration of appropriate primers. Two separate amplifications were done from the working stock of the oligo library for the ON and OFF states, respectively. One amplification, for the ON state, used a primer hybridizing to the 5′ common backbone region. The resulting insert contained both the Switch RNA module and the Trigger attached to its 5′ end. The second amplification, for the OFF state, used a primer hybridizing to the 20 nt unstructured Linker and included a T7 promoter and the 5′ common backbone region in its tail. The OFF-state insert contained only the Switch RNA module without the Trigger module attached. See FIG. 6A for a full schematic of the amplification scheme. A third amplification linearized a ColE1 plasmid backbone for subsequent ligation. This backbone was the same ColE1 backbone as was used in Green et al. (SI-1) for transcribing trigger RNAs, but with a GFPmut3b-ASV gene inserted. All amplicons were cleaned from their reaction buffers by using carboxyl-coated magnetic beads (SI-5) (protocol 4.3): 1× concentration of beads to clean the longer linear backbone product, and 2× bead concentration to clean the smaller insert products. Both inserts were ligated separately into the ColE1 backbone in front of the GFPmut3b-ASV gene using golden gate cloning, as follows. The linearized plasmid backbone was diluted to 500 ng total mass. The ON or OFF insert was added according to a 1:1 molar ratio of insert to plasmid backbone. The inserts and backbone dilutions were prepared into 50 uL ligation reaction volumes, containing 5 uL NEB buffer 3.1, 5 uL T4 ligase buffer, 1 uL BsmBI, 0.5 uL Dpn1, 1 uL T4 ligase, and any remaining volume with nuclease-free water. The 50 uL reaction was placed into a thermocycler for 100 cycles of two steps: 16 C for 10 min and 37 C for 10 min. A final enzyme inactivation step at 65 C for 15 min was done. The ligation products were precipitated out of their reaction buffers using ethanol precipitation. The 50 uL ligation reactions were added to 1.5 mL Eppendorf tubes containing 150 uL of pure ethanol, 5 uL 0.3M sodium acetate (pH 5.2), and 1 uL glycoblue. Tubes were left on dry ice for 20 min and then immediately placed in a 4 C tabletop centrifuge and spun at max RPM for 30 min. Tubes were decanted, and 175 uL of 70% ethanol was added to the tube containing the pellet. Tubes were spun at max speed for 5 min. Tubes were then removed from the centrifuge, decanted, and allowed to dry for 15 min. Ligation products were then eluted in 4 uL TE buffer. For initial library transformation, 50 uL EclonI Supreme cells were given the full 4 uL ligation product elution and electro-transformed. Transformation efficiencies exceeding 10⁷CFU/mL were achieved, and the expanded cells were harvested using a MaxiPrep kit (Qiagen). The resulting pool of plasmids was then electroporated into BL21 star E. coli, where transformation efficiencies exceeding 10⁶were achieved.

Flow-Seq Pipeline

Induction was achieved by expanding BL21 cells overnight at 37 C in LB media with carbenicillin (carb) selection and then diluted 50× into fresh media. After the cells reached OD600 of 0.3 at 37 C (˜2 hours of growth), 0.2 mM IPTG was added, and the cells were allowed to express for another 3 hours at 37 C. The cells were then moved to room temperature and sorted on a Sony SH800 FACS machine with four bins. A positive control consisting of Switch #4 from Green et al. (SI-1), one of the highest performing switches from that study's first-generation design, was cloned both in its OFF state and in the modified fused-trigger ON state. This positive control switch was then used to mark the highest and middle bins of GFP signal, while a negative control consisting of a pUC19 plasmid (containing no GFP) was used to mark the lowest bin of GFP signal (FIG. 7). Approximately 40 million events were sorted for each library. Cells in collected bins were diluted 10× into fresh LB media with carb selection and allowed to expand overnight at 30 C. The expanded cells were then harvested using a MaxiPrep kit (Qiagen).

Deep Sequencing, Read Data Processing and Read Count Analysis

Plasmid collected from sorted cells was amplified using NEB Q5 polymerase 2×MM and primers targeting the common backbone region upstream and downstream of the variable toehold region. The resulting 184 bp (OFF) or 224 bp (ON) PCR products were then analyzed by NGS using a MiSeq or NextSeq instrument (Illumina). Raw paired-end sequencing reads were quality filtered and merged with PEAR 0.9.1. Only sequences matching our intended designs were retained for further analysis. For the ON and OFF libraries, respectively, 10,390,207 reads and 20,788,966 reads were mapped to a correct switch sequence. The individual fluorescence distribution of the ON and OFF state for each switch was measured by calculating its frequency in each bin and assigning a normalized signal metric in the range of [0,1] (FIGS. 1C, D). An ON/OFF metric was calculated as the difference between the ON and OFF signal metrics independently (FIG. 1E). Frequencies of each variant were tabulated for each cell-sorted bin and normalized to the total reads per bin. Each variant's functional value was computed as the weighted mean of its normalized frequencies across all bins, scaled between 0 and 1. The ON/OFF ratios were then calculated as the subtracted difference between ON and OFF (since the fluorescence data had been collected on a logarithmic scale), resulting in a range scaled between −1 and 1.

Library Quality Control

A second biological replicate of our flow-seq pipeline was carried out that produced 60,800 ON measurements, 98,295 OFF measurements, and 30,101 ON/OFF ratio measurements where both ON and OFF were available for the same switch. The R²and MAE between our two datasets were calculated at different read count thresholds. Based on the results (FIGS. 8A-B), five different QC thresholds were established, some of which also included standard deviation cutoffs (Table 1, FIG. 18). QC1 and QC2 contained OFF data with significantly worse R²compared to QC3, QC4, and QC5, but only QC1 contained OFF data with worse MAE. We determined that the inter-replicate drop in R²for OFF values was mainly due to the skewness of the data—indeed, the OFF data consistently showed worse R²values than the ON data throughout the paper, despite having consistently better MAE values. Therefore, we chose to trust in the inter-replicate MAE values more than the inter-replicate R²metric for the OFF data.

To further evaluate the different QC levels, the most stringent data (QC5) were withheld as a test set, and an MLP fed a one-hot representation of the toehold sequence was trained on the four lower QC levels. The results for both predictive R²and MAE showed QC1 to be of significantly inferior quality, but QC2, QC3, and QC4 to be of roughly similar quality (FIG. 9). This result was consistent with the fact that inter-replicate MAE was notably worse at the QC1 count threshold but essentially unchanged across the read count thresholds contained by QC2, QC3, and QC4. The QC2 dataset gave the best predictive results by a small margin and was also significantly larger than QC3 or QC4 (Table 1). With these analyses in mind, QC2 was chosen as the final threshold for inclusion in our dataset. Within the measured ON/OFF ratios in the QC2 dataset, 40,824 had triggers of viral origin, 47,005 had triggers of human origin, and 3,705 had randomly generated trigger sequences.

Cell-Free Switch Validation

Eight of the best switches and eight of the worst switches were synthesized as PCR products, as previously described (SI-2). Briefly, they were ordered as single Ultramer oligos (IDT) without the Trigger fused, from the T7 promoter to the first 36 nt of the common linker and GFP sequences. These were added to a GFP gene by a single PCR amplification step. Triggers were in vitro transcribed from separate oligos that contained the antisense sequence and the antisense T7 promoter, to which the sense strand of the T7 promoter was annealed. Trigger RNA was purified using an RNA Clean & Concentrator kit (Zymo), while Switch DNA was purified using a MinElute kit (Qiagen). To a 5 uL PURExpress reaction were added 2 U/uL Murine RNAse Inh, 5 nM of Toehold Switch PCR product, and either no Trigger RNA or 10 uM of Trigger RNA. Measurements of GFP velocity can be found in FIG. 10. The exact Switches tested and their library assay measurements can be found in Table 2.

Calculations Made with ViennaRNA, Kinfold, and the RBS Calculator

All thermodynamic MFE and ensemble defect calculations, as well as kinetic Kinfold calculations, were obtained using a custom-made python code including libraries from packages such as Biopython (Ref: github.com/biopython/biopython), ViennaRNA (Ref: github.com/ViennaRNA/ViennaRNA), RNAsketch (Ref: github.com/ViennaRNA/RNAsketch) and Pysster (Ref: github.com/budach/pysster). Calculations of thermodynamic rational parameters to include in our database were obtained from toehold RNA sequences by taking each basal 145-nucleotide toehold sequence and then isolating different sections (e.g., GGG, Trigger, Loop1, Switch, Loop2, Stem1, AUG, Stem2, Linker, Post-linker) into distinct sub-sequences with biological relevance for functional analysis (see FIG. 6A, Table 4). Minimum Free Energy (MFE) was calculated for all these sections using the previously reported python-based ViennaRNA Library. MFE calculation using ViennaRNA also specifies a secondary structure in dot-parens-plus notation (unpaired base=dot, base-pair=matching parentheses, and nick between strands=plus). Ideal structures are assumed to be connected and free of pseudoknots. These ideal secondary structures for such sections are:

- SwitchOFF=‘ . . . ((((((((( . . . (((((( . . . )))))) . . . )))))))))’
- SwitchOFF_GFP=‘ . . . ((((((((( . . . (((((( . . . )))))) . . . ))))))))) . . . ((( . . . ((((( . . . ))))) . . . ))) . . . ’
- SwitchOFF_NoTo=‘(((((((( . . . ((((( . . . )))))) . . . ))))))))) . . . ((( . . . ((((( . . . ))))) . . . ))) . . . ’
- SwitchON=‘ . . . (((((((((((((((((((((((((((((( . . . )))))))))))))))))))))))))))))) . . . ’
- SwitchON_GFP=‘ . . . (((((((((((((((((((((((((( . . . )))))))))))))))))))))))))) . . . ((( . . . ((((( . . . ))))) . . . ))) . . . ’
- ToeholdON=‘ . . . (((((((((((( . . . ))))))))))))’
- Stem=‘((((((((( . . . (((((( . . . )))))) . . . )))))))))’
- StemTop=‘(((((( . . . ))))))’

Ensemble defect as a rational parameter was calculated via ViennaRNA/NUPACK for each of the toehold switches in the above subsets of sequence regions: SwitchOFF, SwitchOFF_GFP, Switch_OFF_NoTo, SwitchON, SwitchON_GFP, ToeholdON, Stem, StemTop. This calculation used both the native (calculated from MFE) and the ideal (predefined above) dot-Bracket representation for each sequence to assess the average number of nucleotides that are incorrectly paired at equilibrium. Thirty rational parameters were calculated for each toehold using these methods (fourteen MFE values, eight ideal ensemble defect values, and eight native ensemble defect values).

Kinetic analyses using Kinfold were run from the ViennaRNA package. The OFF-switch sequence was selected, spanning nucleotides 50 to 134 in Table 4 from the start of the toehold to the end of the linker. Due to the large size of the toehold switch RBS, Kinfold trajectories ran for 100-1000× longer than for RBS's previously analyzed relating to the RBS calculator in Borujeni et al. (SI-6) (FIG. 12B). Hence our analysis was scaled down to the QC4 dataset (containing 19,983 total switches), with 100 Kinfold trajectories run for each switch with a maximum stopping time of 10³arbitrary Kinfold units (au). The energy and time at each step of each trajectory were recorded. If the MFE structure was reached within 10³au, it was assumed that the RNA would remain in the MFE structure for the rest of the 10³au timeframe. From each energy trajectory spanning 10³au, the average energy (in kcal/mol) was calculated by integrating the energy-time curve and dividing by 10³. For each switch, the following features were extracted: the mean and standard deviation of the average energy of its 100 sampled trajectories (FIG. 12C), the ratio of the mean average energy to the MFE (FIG. 12E), and the fraction of trajectories that reached the MFE structure within the analyzed 10³timeframe (FIG. 12D).

For predictions by the RBS Calculator, an API was used to access the most recent publicly available version (2.1). Due to limiting computational costs, the QC3 dataset was used instead of the QC2 dataset. For each switch, the translation initiation rate (TIR) of the on-target start codon was predicted for both the ON and OFF states (“SwitchON_GFP” and “SwitchOFF_GFP” respectively in Table 4).

K-Mer Motif Search

In order to compare sequence-level motifs between the best and worst variants measured in our dataset, we performed a k-mer search for over-represented sequence motifs at the tails of our observed functional values. We first filtered the variants for high quality, retaining those with a QC4 score or above. We then took the top and bottom 1,000 variants based on the ON and OFF functional values, respectively. We utilized DREME (SI-7) to test for enrichment or depletion of all possible subsequences of length 3-16 bases, using the indicated foreground and background frequencies. All results above the default E-value cutoff are shown (FIG. 3A, Table 3)

Deep Learning Model Architectures

MLP—Rational Features

The multilayer perceptron (MLP) model based on rational features included a 30-feature input followed by three dense fully connected layers of 25, 10, and 7 neurons, respectively, with rectified linear unit (ReLU) activation, batch normalization, and 20% dropout. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

MLP—OneHot Seq

The MLP model based on the one-hot encoded full 145-nucleotide sequence input was achieved by using a flatten layer followed by three dense layers with ReLU activation, batch normalization, and 20% dropout. Dense layers used 128, 64, and 32 neurons, respectively. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

MLP—Hybrid Rational Features/OneHot Seq

The ensemble MLP model was based on the rational features, as well as a one-hot encoded full 145-nucleotide sequence as input. To construct this model, two networks were assembled in parallel. The first network uses the same architecture for the MLP model with rational features, while the second network used the architecture of the MLP model for one-hot encoded 145-nucleotide sequences. Both networks were then concatenated and connected to a four-neuron dense fully connected layers with ReLU activation. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

CNN—OneHot Seq

The Convolutional Neural Network (CNN) model based on the one-hot encoded full 145-nucleotide sequence as input was achieved by direct feeding of the input to three convolutional layers with ReLU activation, batch normalization, and 20% dropout. The convolutional layers used had 32, 64, and 128 filters of size 3, respectively. Same-padding was used with L1 and L2 kernel regularization. The output from the convolutional layers was flattened and fed to two fully connected sequential dense layers of 16 neurons each with ReLU activation, batch normalization, and 20% dropout. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

CNN—2D Complementarity Map

The Convolutional Neural Network (CNN) model based on the one-hot encoded categorical 2D complementarity-directional matrix from the full 145-nucleotide sequence as input was achieved by direct feeding of the input to three convolutional layers with ReLU activation, batch normalization, and 30% dropout. The convolutional layers used had 32, 64, and 128 filters of size 5×5 respectively. Same-padding was used with L1 and L2 kernel regularization. The output from the convolutional layers was flattened and fed to two fully connected sequential dense layers of 16 neurons each with ReLU activation, batch normalization, and 20% dropout. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

LSTM—OneHot Seq

The Long Short-Term Memory (LSTM) recurrent neural network model on the one-hot encoded full 145-nucleotide sequence as input was achieved by direct feeding of the input to a network with 128 recurrent units. The output of this was then connected to 100-neuron fully connected dense layer with ReLU activation, followed by batch normalization and 30% dropout. This network was then fed to a final three-neuron layer (ON, OFF, ON/OFF) with linear activation for regression output, or to a final two-neuron layer (ON/OFF: binarized at +/−0.7) with softmax activation for classification output.

All models were trained using a maximum of 300 epochs, considering a 20-epoch early stopping patience, which gets triggered upon lack of model improvement on the validation set. Batch size for all models was 64*(1+ngpus), where ngpus is defined as the number of used graphic processing units during model training. All trained regression models were verified for reported metrics using 10-fold cross-validation, while classification-trained models were evaluated on three shuffled test sets as indicated.

Complementarity Matrix and VIS4Map

Complementary maps were defined as a One-Hot Encoded Categorical 2D Complementarity-directional Matrix (total number of tensor dimensions=3) constructed by defining columns and rows of the matrix as the position of potential complementarity between any two given pairs of nucleotides in a single RNA sequence. The value in each position is defined as a one-hot encoded categorical variable according to the Watson-Crick pairing of the two nucleotides defining that position. Nucleotide pairings are assigned the following category: G-C (6)=[0 0 0 0 0 1], C-G (5)=[0 0 0 0 0 1 0], A-U (4)=[0 0 0 0 1 0 0], U-A (3)=[0 0 0 1 0 0 0], G-U (2)=[0 0 1 0 0 0 0], U-G (1)=[0 1 0 0 0 0 0], NonWCpairs (0)=[1 0 0 0 0 0 0]. VIS4Maps were generated using a modified algorithm, attention, activation maximization and saliency map visualization for Keras (Keras-Vis, Ref: github.com/raghakot/keras-vis) with tensorflow backend.

In this case, gradients were calculated from a regression model for all regions of the image to visualize what spatial features cause the predicted output to increase. To visualize the toehold regions that are mostly responsible for each prediction, small positive or negative gradients are highlighted using a normalization strategy. Given this information, such techniques allow us to generate heatmap-encoded saliency map images that spatially relate to the toehold regions in the complementarity map that lead to accurate predictions.

Results

Engineered ribonucleic acid (RNA) molecules with targeted biological functions play an important role in synthetic biology (1), particularly as programmable response elements for small molecules, proteins, and nucleic acids. Examples include riboswitches, riboregulators, and ribozymes, many of which hold great promise for a variety of in vitro and in vivo applications (1, 2). Despite their appeal, the design and validation of this emerging class of synthetic biology modules have proven challenging due to variability in function that remains difficult to predict (2-9). Current efforts aiming to unveil fundamental relationships between RNA sequence, structure, and behavior focus mostly on mechanistic thermodynamic modeling and low-throughput experimentation, which often fail to deliver sufficiently predictive and actionable information to aid in the design of complex RNA tools (2-9). Deep learning, by contrast, constitutes a set of computational techniques well suited for pattern recognition in complex and highly combinatorial biological problems (10-14), such as the sequence design space of RNA tools. However, the application of deep learning to predicting function in RNA synthetic biology has been limited by a notable scarcity of datasets large enough to effectively train deep neural networks. Toehold switches, in particular, represent a canonical RNA element in synthetic biology that could greatly benefit from deep learning approaches to better predict function and elucidate useful design rules.

Toehold switches are a class of versatile prokaryotic riboregulators inducible by the presence of a fully programmable trans-RNA trigger sequence (2-6, 15, 16). These RNA synthetic biology modules have displayed impressive dynamic range and orthogonality when used both in vivo as genetic circuit components (2, 5, 6), and in vitro as nucleic acid diagnostic tools using cell-free protein synthesis (CFPS) systems (3, 4, 15, 16). Similar to other RNA synthetic biology tools, a substantial fraction of toehold switches show poor to no measurable function when tested experimentally, and while efforts have been made to establish rational, mechanistic rules for improved performance based on low-throughput datasets (2-9, 15, 16), the practical utility of these approaches remains inconclusive. Thus, considering the wide applicability and general challenges of toehold switch design, our objective in this study was to develop a deep learning platform to predict toehold switch function as a canonical RNA switch model in synthetic biology.

To achieve this goal, we first aimed to expand the size of available toehold datasets using a high-throughput DNA synthesis and sequencing pipeline to characterize over 10⁵new toehold switches. We then used this comprehensive new dataset to demonstrate that deep neural networks trained directly on switch RNA sequences can outperform rational thermodynamic and kinetic analyses to predict toehold switch function. Furthermore, we enhanced the transparency of our deep learning approach by utilizing a nucleotide (nt) complementarity matrix input representation to visualize learned secondary structure patterns in selected models. This attention-visualization technique, which we term VIS4Map (Visualizing Secondary Structure Saliency Maps), allowed us to identify RNA module failure modes by discovering secondary structures that our deep learning model used to accurately predict toehold switch function. The resulting dataset, models, and visualization analysis (FIG. 1) represent a substantial step forward for the validation and interpretability of high-throughput approaches to designing RNA synthetic biology tools, surpassing the limits of current mechanistic RNA secondary structure modeling.

Library Synthesis, Characterization, and Validation

As mentioned previously, a fundamental hurdle in applying deep learning techniques to RNA synthetic biology systems is the limited size of currently published datasets, which are notably smaller than typical dataset sizes required for training of deep network architectures in other fields (10, 17-21). For example, to date, less than 1000 total toehold switches have been designed and tested (2-6, 9, 15, 16), a situation that currently limits the synthetic biology community's ability to utilize deep learning techniques for analysis of this type of response molecules. Therefore, towards improving our understanding and ability to predict new functional RNA-based response elements, we first set out to synthesize and characterize an extensive in vivo library of toehold switches using a high-throughput flow-seq pipeline (22) for subsequent exploration using various machine learning and deep learning architectures.

Our toehold switch library was designed and synthesized based on a large collection (244,000) of putative trigger sequences, spanning the complete genomes of 23 pathogenic viruses, the entire coding regions of 906 human transcription factors, and ˜10,000 random sequences. From a synthesized oligo pool, we generated two construct libraries, for ON and OFF states, which were subsequently transformed into BL21 Escherichia coli (FIGS. 1, 6A-B). The first library contained OFF toehold switch constructs that lacked a trigger, while the second library of ON constructs contained the same toeholds with the complementary triggers fused to their corresponding switches. The two libraries were then sorted on a fluorescence-activated cell sorter (FACS) using four bins (FIGS. 1, 7), and the toehold switch variants contained in each bin were quantified using next-generation sequencing (NGS) to recover their individual fluorescence distributions from raw read counts (FIG. 1). After quality control (Table 1), the toehold switch library contained 109,067 ON state measurements (FIG. 2A), 163,967 OFF state measurements (FIG. 2B), and 91,534 ON/OFF paired difference ratios (FIG. 2C), where both ON and OFF states were characterized for each switch (FIGS. 2E,F). ON and OFF data were normalized from 0 to 1, resulting in an ON/OFF difference ratio normalized from −1 to 1.

Since RNA synthetic biology tools such as toehold switches are often used within in vitro cell-free systems (3, 4, 15, 16), we validated our in vivo ON/OFF measurements in an in vitro setting to ensure these were reasonable indicators of switch performance in a CFPS system. To achieve this, we selected eight high-performance switches and eight low-performance switches, and individually cloned and characterized them in a PURExpress CFPS (FIGS. 1D, 10 and Table 2). All low-performance switches showed no induction, while the high-performance switches showed a spread of cell-free ON/OFF ratios between 2 and 10 (p<0.0001 between high and low switches, two-tailed t-test). These results confirm that while the performance of toehold switches in vivo and in vitro may differ, in vivo measurements can still be used to classify categorically whether a switch will function in vitro.

Rational Analysis Using Thermodynamic RNA Secondary Structure Models

Before initiating the exploration of deep learning models to predict function in our large-scale toehold switch library, we sought to determine whether traditional tools for analyzing synthetic RNA modules could be used to accurately predict toehold switch behavior, including k-mer searches and mechanistic modeling using thermodynamic parameters. K-mer searches of biological sequence data are often used to discover motifs, and while certain overrepresented motifs were found in our dataset (FIG. 3A and Table 3), utilization of these did not significantly improve functional predictions of switch behavior. Other current state-of-the-art approaches for designing RNA synthetic biology tools primarily analyze secondary structure using thermodynamic principles (23-25). Following such prior works, we used NUPACK (23) and ViennaRNA (25) software packages to calculate a total of 30 rational features for our entire library, including the minimum free energy (MFE), ideal ensemble defect (IED), and native ensemble defect (NED) of the entire toehold switch library as well as various sub-segments in each sequence (Table 4). A number of these parameters had previously been reported to correlate with experimental toehold switch ON/OFF measurements for smaller datasets (2), and NUPACK's design algorithm, in particular, is set to optimize IED when proposing target RNA secondary structures (3, 4, 15, 23). However, when analyzing these rational features with our larger dataset, we found them to be poor predictors of toehold switch function (FIGS. 3B, 11A-B). In modest agreement with the findings of Green et al. (2), the MFE of the RBS-linker region showed the highest correlation of this feature set for ON/OFF (R²: ON=0.14, OFF=0.06, ON/OFF=0.04), with NUPACK's IED also showing above-average correlation (R²: ON=0.07, OFF=0.02, ON/OFF=0.03). While measurable, these correlation metrics were far too weak for practical use in computer-aided design of this specific RNA synthetic biology tool (3, 4, 15, 23).

Moving forward, we explored the use of more complex thermodynamic models that take into account well-established hypotheses for translation initiation and the ribosome docking mechanism in combination with multiple thermodynamic features to improve their predictions (26-31). One of the most developed of these models is the Ribosome Binding Site (RBS) calculator (v2.1; Salis Lab); a comprehensive model parameterized on thousands of curated RBS variants (26-29). We used the RBS calculator to predict the ON and OFF translation initiation rates for our toehold switches, but also found low predictive performance comparable to other rational features (FIG. 3B) when tested on our database (R²: ON=0.09, OFF=0.05, ON/OFF=0.0001).

One potential explanation for the limited predictive power of current thermodynamic models for RNA folding tasks concerns the influence of kinetically stable secondary structure intermediates that may compete with thermodynamic equilibrium states (29, 32). To determine whether a kinetic analysis of toehold switch folding dynamics could help explain our experimental results, we calculated four additional features based on kinetic trajectories using the Kinfold package (33) (FIG. 12A-F). As with predictions obtained using other thermodynamic models, these kinetic features showed poor correlations (R²: ON=0.04, OFF=0.04, ON/OFF=0.001 for the best feature) to our empirical dataset (FIG. 12E). Considering these results, the cause of limited function predictions from thermodynamic and kinetic models of RNA secondary structure remains unclear but may stem from the use of potentially incomplete energetic models, incorrect mechanistic hypotheses, and from interference within the in vivo context of the bacterial cells. Regardless of the source of error, we sought to explore deep learning as a machine learning paradigm to develop models with higher predictive abilities than previously reported, with the hope of allowing useful computer-aided systems for the design of RNA synthetic biology tools.

Improved Prediction Using Sequence-Based Multilayer Perceptron Models

Given that simple regression models based on state-of-the-art RNA thermodynamic and kinetic calculations were ineffective at predicting toehold switch performance, we next tested the use of a type of feed-forward neural networks, also known as multilayer perceptron (MLP) models, as a baseline architecture for our investigation (FIG. 3C). We first trained a three-layer MLP model on our dataset with an input consisting of the 30 previously calculated thermodynamic rational features (see Methods section for further detail). When trained in regression-mode, this MLP model was able to deliver better predictions than any of the individual rational features or the RBS calculator based on R²and MAE (R²: ON=0.35, OFF=0.25, ON/OFF=0.20) (FIGS. 3D, E). Similarly, when this model was trained for classification (ON/OFF: binarized at +/−0.7), as seen in FIG. 13, it achieved a 0.76 area under the receiver-operator curve (AUROC) and 0.18 area under the precision-recall curve (AUPRC), as seen in FIG. 3F. The MLP model modestly outperformed a logistic regressor trained on the same rational features (FIGS. 3D,E,F), suggesting that the MLP architecture was able to abstract higher-order patterns from these features as compared to simpler non-hierarchical models.

While these results already constitute an improvement compared to the current state-of-the-art analysis of RNA synthetic biology tools, we wondered whether the use of pre-computed rational features as network input led to information loss that could inherently limit the predictive power of these models. Considering that possibility, we trained an MLP model solely on one-hot encoded sequence representations of our toehold switches, eliminating potential bias introduced by a priori mechanistic modeling. We found that this sequence-based MLP delivered improved functional predictions based on R²and MAE (R²: ON=0.70, OFF=0.53, ON/OFF=0.43) metrics (FIGS. 3D, E, 14). These values represent a doubling of R²performance as compared to the MLP trained on rational features and a ten-fold improvement in ON/OFF R²over the best rational feature used for previous linear models.

Similarly, when training for classification, our one-hot sequence MLP produced improved AUROCs and AUPRCs, reaching 0.87 and 0.36, respectively (FIG. 3F). This improvement in performance when training on sequence-only inputs compared to rational features suggests that significant information loss occurs when performing thermodynamic calculations on toehold switch sequences, a problem that may extend to other RNA synthetic biology tools in use today. Also, the sequence-only MLP model dramatically outperformed the logistic regressor model trained on the same one-hot sequence input (FIGS. 3D,E,F), further supporting the hypothesis that improved accuracy of our sequence-based MLP arises from learned hierarchical non-linear features extracted directly from RNA sequences. A combined input concatenating both the rational features and the one-hot representation was also evaluated, giving a small, but significant improvement in regression mode (ΔR²≈0.025 and ΔMAE≈−0.0025, p<0.05 for all six comparisons, two-tailed t-test), but with no significant improvement for AUROC or AUPRC when in classification mode (FIGS. 3D,E,F). These results suggest that while the use of rational features may facilitate the abstraction of potentially relevant information of toehold switch function, the one-hot sequence-only MLP model can recover such information if given enough training data without a priori hypothesis-driven assumptions built into the model.

In order to validate the degree of biological generalization in our sequence-only MLP model, we withheld 23 viral genomes tiled in the toehold switch dataset during training and predicted their function resulting in a 0.82-0.98 AUROC range (average 0.87, FIG. 15). We then carried out an external validation on a previously published dataset of 168 characterized toehold switches (2) that our models had never seen before and that had been collected under different experimental conditions. Our MLP models achieved an AUROC of 0.70, 0.81, and 0.79, when trained on rational features, one-hot sequence, and hybrid inputs, respectively (FIG. 3G). The improved performance observed when training the models directly on nucleotide sequence rather than thermodynamic features, even for an external dataset, suggest a competent degree of biological generalization and supports the value of modeling RNA synthetic biology tools using deep learning and high-throughput datasets, removing current assumptions from mechanistic rational parameters.

Predictive Performance of Higher-Capacity Deep Learning Models

Having explored relatively simple deep learning architectures first, we next sought to determine whether training our dataset on higher-capacity convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent neural networks could increase our predictive ability. CNN and LSTM models have been applied to a variety of biological datasets in recent years, and have been cited as being particularly adept at recognizing motifs and long-range interactions in nucleotide sequence data (10, 17-20, 34-38). We specifically evaluated a CNN trained on a one-hot sequence input, an LSTM trained on a one-hot sequence input, and a CNN applied to a two-dimensional (2D), one-hot complementarity map representation input (see Methods for complete descriptions of all models). Upon evaluating both the R²and MAE in regression mode and the AUROC and AUPRC in classification mode for these models (FIGS. 4A,B,C,D), we concluded that these neural network architectures did not lead to superior predictive models, as compared to the sequence-based, three-layer MLP described previously. We reasoned that, in these specific cases, increased model capacity led to under- or over-fitting, requiring additional training examples or improved fine-tuning to accelerate effective training.

Visualizing Learned RNA Secondary Structure Motifs with VIS4Map

One significant drawback of using deep learning approaches to predict biological function is the inherent difficulty in understanding learned patterns in a way that aids researchers in elucidating biological mechanisms underlying the model predictions. By contrast, mechanistic hypothesis-driven models can more directly inform which aspects of a biological theory best explain the observations. Various methods have been established to address this limitation, including alternative network architectures (39), and the use of saliency maps (40, 41), which reveal the regions of an input that deep learning models pay attention to when making predictions. While saliency maps have been previously used to visualize model attention in one-hot representations of sequence data (10, 17, 18, 20, 40), such implementations focus only on the primary sequence and have not been developed to identify secondary structure interactions, which are specially relevant in the operation of RNA synthetic biology elements. In the few cases where secondary structure has been investigated, input representations have been constrained to predetermined structures based on the predictions of thermodynamic models (37, 38), whose abstractions we have found cause significant information loss.

In order to better explain our deep learning model's predictions, we sought to visualize RNA secondary structures learned by our neural networks in a manner unconstrained by thermodynamic modeling. To achieve this, we chose to use a CNN trained on two-dimensional nucleotide complementarity map representations (FIG. 5A) to allow for attention pattern visualization in this secondary structure space. Each position in this complementarity map corresponds to the potential pair between two nucleotides, indicating its identity with a one-hot encoding (G-C, C-G, A-U, U-A, G-U, U-G, or an unproductive pair). We hypothesized that by training deep networks on such a representation of RNA sequences, it would be possible for generated saliency maps to reveal learned secondary structure features as visually intuitive diagonal features. Importantly, because the complementarity map is unconstrained by a priori hypotheses of RNA folding (similarly to our sequence-based MLP models), we anticipated this approach to be able to identify secondary structures that might be overlooked by commonly used thermodynamic and kinetic algorithms, such as NUPACK and Kinfold.

To validate the feasibility of our visualization approach, we first pre-trained a CNN to predict NUPACK MFE values from complementarity map representations of a randomly selected in silico RNA sequence dataset. Because MFE is directly determined by RNA secondary structure, we anticipated that a CNN undergoing this pre-training would likely pay attention to secondary structure features, a situation that was confirmed through visualization of individual attention maps (FIGS. 5B,C). Additionally, we found that the use of a complementarity map input improved the CNN's predictions of MFE from R²=0.6 to R²=0.74 compared with a one-hot sequence input (FIG. 16). Indeed, the saliency maps generated from a CNN trained on a complementarity map input contained primarily diagonal features that showed a statistically significant degree of agreement with the MFE structures from which NUPACK based its MFE calculations (FIGS. 5B, C, 16). Hence without prior knowledge of the algorithm or parameters NUPACK uses to calculate MFE, our CNN was able to learn similar abstractions as NUPACK, which we then used to intuitively visualize underlying relevant RNA secondary structures utilizing our complementarity map input representation. We named this approach for interpreting RNA deep learning models Visualizing Secondary Structure Saliency Maps or VIS4Map.

Encouraged by our CNN's ability to elucidate RNA secondary structure features directly from training data, we applied VIS4Map to our entire toehold switch dataset. When trained on a complementarity map representation of the switch OFF conformation (FIG. 5D) both in regression mode and classification mode, VIS4Map significantly outperformed an MLP trained on rational thermodynamic features; however, VIS4Map did not significantly outperform our MLP trained on one-hot inputs, similar to the case of our other higher capacity models (FIGS. 4A,B,C,D). Encouragingly, nonetheless, we found that saliency maps produced by this CNN model displayed clear diagonal secondary structure features (FIG. 5D). These structures appear to span from hybridization between the toehold and the ascending stem, to hybridization between the descending stem and the downstream linker. We confirmed the biological relevance of these features by averaging saliency maps and found that the shared structures corresponded to the designed on-target structure of the switch hairpin (FIG. 5E). We further analyzed learned features outside of the designed MFE structure by sorting saliency maps using the toehold switch OFF signal (FIGS. 5F and 17). We found that for leakier (high OFF) switches, the CNN identified a high degree of salient off-target secondary structures that could compete with the main hairpin stem and thereby exposed the RBS, whereas for tight (low OFF) switches the CNN identified fewer competing off-target secondary structures. In the context of general riboregulator behavior, these findings support the hypothesis that leaky expression from an RBS repressed by secondary structures can be caused by the misfolding of the repressive structure into less stable kinetic intermediate conformations (29, 32) (FIG. 5F, right).

The fact that VIS4Map was able to identify both equilibrium and kinetically stable RNA secondary structures indicates a remarkable ability to uncover biologically relevant information, which in this case supports currently postulated hypotheses on prokaryotic translation initiation. Importantly, the identified secondary structure features could not have been visualized using the one-hot sequence representation commonly associated with saliency maps (10, 17, 18, 20). These findings compound to the advantage of using sequence-only deep learning approaches for analyzing RNA synthetic biology tools. Outside of toehold switches and other synthetic RNA systems, we anticipate VIS4Map will be broadly useful for the discovery of previously unknown equilibrium or kinetically stable structures contributing to RNA biology, that are not predicted by current mechanistic RNA structure models.

Discussion

Here we presented a high-throughput DNA synthesis, sequencing, and deep learning pipeline for the design and analysis of a synthetic system in RNA biology. Having produced a toehold switch dataset ˜100-fold larger than previously published as a model system for investigating synthetic RNA response elements (2-6, 15, 16), we demonstrated the benefits of using deep learning methods that directly analyze sequence rather than relying on calculations from mechanistic thermodynamic and kinetic models. This approach resulted in tenfold improvement in functional prediction R²over an ensemble of commonly used thermodynamic and kinetic features. Moreover, the validation of our deep learning models on an external previously characterized dataset, as well as the holdout prediction of every individual viral genome in our dataset, further demonstrated the robust biological generalization of our models.

As with most work in RNA synthetic biology, all previous attempts to improve toehold switch functionality have relied on the guidance of mechanistic thermodynamic modeling and low-throughput datasets (2-8, 15, 16). Too frequently, rational design rules fail to give meaningful predictions of function for RNA-based synthetic systems. The results presented here suggest that the biological processes underlying RNA biology may be more complex than current state-of-the-art analyses take into account and that high-throughput DNA synthesis, sequencing, and deep learning pipelines can be more effective for modeling said complexity. Combining improved predictions with enhanced understanding, our novel VIS4Map method further allowed us to visualize the equilibrium and kinetic secondary structure features that our deep learning models identified as important to the leakage of the switch OFF state. While secondary structures identified by NUPACK, Kinfold, and other rational mechanistic models are limited by predefined abstractions, which may cause significant information loss, our approach explored sequence space in an unrestricted manner and analyzed all possible RNA secondary structures. VIS4Map could prove useful for identifying complex secondary structure information that might otherwise be ignored by simplified physical energetic models of RNA folding.

The dataset reported here also represents an extensive repository of characterized toehold switches, which could be used to accelerate the development of future cell-free diagnostics (3, 4, 15, 16). These switches tile the entire genomes of 23 pathogenic viruses of high clinical importance, as well as tiling hundreds of human transcripts, including many that are differentially expressed in cancerous phenotypes (42, 43). The total cost of our flow-seq pipeline equates to ˜$0.08 per measurement, suggesting that the benefits of high-throughput design and assaying of RNA synthetic biology tools could be made widely accessible. We hope that this work will encourage the use of high-throughput data collection for the training of deep learning systems, paired with more interpretable neural network architectures unrestricted by thermodynamic or kinetic secondary structure models for improved prediction and insight generation in RNA synthetic biology.

TABLE 1

Quality control thresholds.

Quality Control Conditions
Library Size

OFF Count
ON Count
Upper Stdev.
Lower Stdev.
ON
OFF
ON/OFF

Threshold
Threshold
Cutoff
Cutoff
Variants
Variants
Variants

QC1
>=5
>=5
None
None
126,620
180,552
110,931

QC2
>=10
>=10
None
None
109,067
163,967
91,534

QC3
>=20
>=40
None
>0
77,040
90,264
43,044

QC4
>=60
>=60
0.4>
>0.04
39,283
67,507
19,983

QC5
>=300
>=300
0.4>
>0.04
6,187
12,551
1,137

The conditions for inclusion in our five quality control groups (QC1-5) are shown above, including standard deviation cutoffs and library count thresholds. QC2 was ultimately chosen as the final condition for inclusion in our dataset, and all data used or shown in this manuscript is for QC2 unless otherwise stated. The size of each dataset is shown in the three rightmost columns.

TABLE 2

Toehold switch sequences validated in cell-free format.

Corresponding

Riboregulator

SEQ ID NO #
Trigger Sequence
On
Off

Low 1
52454
CCGACACCTGTTTCATGGAACAATAAAAGA
0.0153
0.0085

(SEQ ID NO: 244004)

Low 2
65651
TGCTGTCTGTGAAACAGATAAATGGAAATA
0.0176
0.0100

(SEQ ID NO: 244005)

Low 3
67027
TCCCTTTCCCAGAAATAAACTTTTTTACCC
0.0181
0.0136

(SEQ ID NO: 244006)

Low 4
235901
TCACTGAGTCATTGCCATCTGCAGAATCAG
0.0048
0.0134

(SEQ ID NO: 244007)

Low 5
81973
TCCAAGACCCAAAGTTCTGGGAACTGGTGG
0.0192
0.0156

(SEQ ID NO: 244008)

Low 6
45174
TGGCAATTGTAGATATAACTTCTGGTAAAT
0.0153
0.0183

(SEQ ID NO: 244009)

Low 7
74706
ATCCAAATATAATGATGACCTATATGCCCT
0.0158
0.0102

(SEQ ID NO: 244010)

Low 8
66097
CCAATATGAGATCTGTAATGCTAACAGTTT
0.0076
0.0146

(SEQ ID NO: 244011)

High 1
19367
GTCATATAAAGGAAGAAGATAGGAGAAGAA
0.9860
0.0031

(SEQ ID NO: 244012)

High 2
236638
AGTTCACAAGAGATGGTTCATGGTGTTCCA
0.9937
0.0132

(SEQ ID NO: 244013)

High 3
111698
AAAGGTTAGCTTATGTTACATATCAAGATA
0.9740
0.0016

(SEQ ID NO: 244014)

High 4
62866
AATCACTGAAAATTGGAGTTAGGTATTGAC
0.9747
0.0007

(SEQ ID NO: 244015)

High 5
40182
GGTATGTTAAGTATGAGGCCTTATCCGTAC
0.9895
0.0115

(SEQ ID NO: 244016)

High 6
9602
TCAAGTTAGAGAAGGAAGTGGCTGAGACCC
0.9856
0.0122

(SEQ ID NO: 244017)

High 7
43841
TAAATCTATGAGAGATCAACGAAAAGGAAG
0.9942
0.0150

(SEQ ID NO: 244018)

High 8
164989
AAAGAAGAAATCATGCAAGAAAACAAAGGG
0.9744
0.0007

(SEQ ID NO: 244019)

Sequences of the individually cloned toehold switches for cell-free validation using PURExpress were selected from the QC3 threshold. Their trigger sequences and flow-seq assay performances are shown (see FIGS. 1F,9 for cell-free assay performance). All highly-functional switches have ON/OFF of 0.97 or greater, while all poorly-functional switches have ON/OFF of 0.04 or less.

TABLE 3

K-mer search results.

Counts in
Counts in

Motif
Foreground
Background
P-value
E-value

ON Triggers

Low versus High Signal

UCUYU
349
0
7.10E-122
8.30E-117

CU*

GAUGG
260
19
6.80E-63
7.90E-58

AAAAA
391
128
1.90E-42
2.10E-37

CUCYU
142
4
1.30E-39
1.40E-34

C*

UAUUA
123
0
1.70E-39
1.90E-34

AC

UCUCA
26
2
4.10E-37
4.50E-32

C*

GAGUC
100
0
5.80E-32
6.30E-27

GU

GUUUU
100
2
8.50E-29
9.10E-24

AUC

High versus Low Signal

ANSA
785
427
6.00E-62
1.00E-56

AWUB
644
359
9.50E-38
7.80E-33

UAYR
355
163
3.90E-23
1.70E-18

GVRA
270
128
8.20E-16
2.50E-11

ACK
344
224
1.60E-09
3.80E-05

AUAA
104
47
8.30E-07
1.40E-02

OFF Triggers

Low versus High Signal

CNG
762
503
8.40E-34
1.50E-28

GRS
510
342
1.90E-14
1.80E-09

CCUH
218
132
2.60E-07
1.60E-02

High versus Low Signal

AWWWU
591
346
2.10E-28
3.60E-23

WUAW
472
333
1.40E-10
1.60E-05

AAAAR
67
22
5.60E-07
4.30E-02

A

K-mer motifs searched with DREME using the trigger RNA sequences of the highest and lowest performing 1000 switches sorted by either ON or OFF signal. For this search, QC3 dataset was selected. * Denotes potential anti-SD pyrimidine-rich sequences.

TABLE 4

Rational feature sub-sequences.

Rational Feature
Sequence

Sub-sequence Name
Region
Brief Description

SwitchOFF
30-108
Toehold switch off conformation

SwitchOFF-GFP
30-144
Off conformation with added GFP sequence

SwitchOFF-NoTo
62-144
Off conformation with toehold removed

SwitchON
0-108
Toehold switch on conformation

SwitchON-GFP
0-144
On conformation with added GFP sequence

Trigger
0-29
Trigger sequence alone

ToeholdOFF
30-62
Toehold region of switch including link1

ToeholdON
0-62
Toehold region only hybridized to trigger

Stem
62-108
Stem only of toehold switch

AscendingStem
62-100
Ascending arm of the switch stem

DescendingStem
80-108
Descending arm of the switch stem

StemTop
74-97
Top half of the stem from start codon up

RBS-Linker
80-134
Region from RBS loop2 to linker

RBS-GFP
80-144
RBS-Linker with added GFP sequence

[−3, −1]
[0, 29]
[30, 49]
[50, 79]
[80, 90]
[91, 96]
[97, 99]
[100, 108]
[109, 134]
[135, 144]

GGG
trigger
loop1
switch
loop2
stem1
AUG
stem2
linker
post-linker

The sub-sequences from which the thirty rational features used as MLP input were calculated using ViennaRNA are shown here in the upper panel. In the lower panel, we show the full un-truncated toehold switch sequence framework from which the sub-sequences in the top table were selected.

TABLE 5

List of SEQ ID NOs: 1-244,000 denoting viral

or transcription factor specificity

Virus/Source Gene

SEQ ID NOs: 1-159175

astrovirus
SEQ ID NO: 1-1298

cardiovirus
SEQ ID NO: 1299-2885

chikungunya
SEQ ID NO: 2886-5245

cosavirus
SEQ ID NO: 5246-6682

coxsackie
SEQ ID NO: 6683-8156

dengue
SEQ ID NO: 8157-10298

ebola
SEQ ID NO: 10299-14080

hantavirus
SEQ ID NO: 14081-14798

human immunodeficiency
SEQ ID NO: 14799-16864

human parvo
SEQ ID NO: 16865-17978

human rhino
SEQ ID NO: 17979-19403

influenza: h1n1
SEQ ID NO: 19404-19603

influenza: h3n2
SEQ ID NO: 19604-20044

lassa
SEQ ID NO: 20045-20719

leishmania
SEQ ID NO: 20720-21752

marburg
SEQ ID NO: 21753-25569

papilloma
SEQ ID NO: 25570-27118

poliovirus
SEQ ID NO: 27119-28601

rabies
SEQ ID NO: 28602-30976

random_sequences
SEQ ID NO: 30977-41790

smallpox
SEQ ID NO: 41791-78900

west nile
SEQ ID NO: 78901-81100

yellow fever
SEQ ID NO: 81101-83267

zika
SEQ ID NO: 83268-85423

human_AC009336.2
SEQ ID NO: 85424-85465

human_AC012531.2
SEQ ID NO: 85466-85501

human_AC097634.4
SEQ ID NO: 85502-85702

human_ACTB
SEQ ID NO: 85703-85812

human_ACTL6A
SEQ ID NO: 85813-85926

human_ACTN4
SEQ ID NO: 85927-86197

human_AEBP1
SEQ ID NO: 86198-86542

human_AEBP2
SEQ ID NO: 86543-86695

human_AGO1
SEQ ID NO: 86696-86950

human_AGO2
SEQ ID NO: 86951-87206

human_AHR
SEQ ID NO: 87207-87458

human_AIRE
SEQ ID NO: 87459-87619

human_AKNA
SEQ ID NO: 87620-87887

human_AL121581.1
SEQ ID NO: 87888-88134

human_ALX1
SEQ ID NO: 88135-88230

human_ALX4
SEQ ID NO: 88231-88351

human_ANHX
SEQ ID NO: 88352-88463

human_AR
SEQ ID NO: 88464-88737

human_ARHGAP35
SEQ ID NO: 88738-89185

human_ARID3A
SEQ ID NO: 89186-89361

human_ARID3B
SEQ ID NO: 89362-89527

human_ARID3C
SEQ ID NO: 89528-89648

human_ARID4A
SEQ ID NO: 89649-90002

human_ARID4B
SEQ ID NO: 90003-90393

human_ARID5A
SEQ ID NO: 90394-90569

human_ARID5B
SEQ ID NO: 90570-90923

human_ARNT
SEQ ID NO: 90924-91157

human_ARNT2
SEQ ID NO: 91158-91370

human_ARNTL
SEQ ID NO: 91371-91556

human_ARNTL2
SEQ ID NO: 91557-91730

human_ARRB1
SEQ ID NO: 91731-91851

human_ARX
SEQ ID NO: 91852-92017

human_ASCL1
SEQ ID NO: 92018-92086

human_ASCL2
SEQ ID NO: 92087-92142

human_ASCL3
SEQ ID NO: 92143-92194

human_ASCL4
SEQ ID NO: 92195-92244

human_ASCL5
SEQ ID NO: 92245-92304

human_ASH2L
SEQ ID NO: 92305-92490

human_ATF1
SEQ ID NO: 92491-92569

human_ATF2
SEQ ID NO: 92570-92718

human_ATF3
SEQ ID NO: 92719-92770

human_ATF4
SEQ ID NO: 92771-92873

human_ATF5
SEQ ID NO: 92874-92955

human_ATF6
SEQ ID NO: 92956-93154

human_ATF6B
SEQ ID NO: 93155-93363

human_ATMIN
SEQ ID NO: 93364-93608

human_ATOH1
SEQ ID NO: 93609-93712

human_ATOH8
SEQ ID NO: 93713-93806

human_ATXN3
SEQ ID NO: 93807-93912

human_BACH1
SEQ ID NO: 93913-94131

human_BACH2
SEQ ID NO: 94132-94381

human_BARHL1
SEQ ID NO: 94382-94477

human_BARHL2
SEQ ID NO: 94478-94591

human_BARX1
SEQ ID NO: 94592-94665

human_BARX2
SEQ ID NO: 94666-94747

human_BASP1
SEQ ID NO: 94748-94813

human_BATF
SEQ ID NO: 94814-94848

human_BATF2
SEQ ID NO: 94849-94928

human_BATF3
SEQ ID NO: 94929-94964

human_BAZ2A
SEQ ID NO: 94965-95533

human_BCL11A
SEQ ID NO: 95534-95771

human_BCL11B
SEQ ID NO: 95772-96037

human_BCL6
SEQ ID NO: 96038-96247

human_BCL6B
SEQ ID NO: 96248-96389

human_BCOR
SEQ ID NO: 96390-96903

human_BHLHA15
SEQ ID NO: 96904-96958

human_BHLHE40
SEQ ID NO: 96959-97079

human_BHLHE41
SEQ ID NO: 97080-97221

human_BORCS8-MEF2B
SEQ ID NO: 97222-97328

human_BRCA1
SEQ ID NO: 97329-97542

human_BRD7
SEQ ID NO: 97543-97735

human_BRF2
SEQ ID NO: 97736-97859

human_CALCOCO1
SEQ ID NO: 97860-98064

human_CARF
SEQ ID NO: 98065-98279

human_CARM1
SEQ ID NO: 98280-98459

human_CBX4
SEQ ID NO: 98460-98625

human_CC2D1A
SEQ ID NO: 98626-98908

human_CC2D1B
SEQ ID NO: 98909-99163

human_CCAR1
SEQ ID NO: 99164-99506

human_CCNT1
SEQ ID NO: 99507-99722

human_CDC5L
SEQ ID NO: 99723-99960

human_CDK12
SEQ ID NO: 99961-100405

human_CDK13
SEQ ID NO: 100406-100856

human_CDK5RAP2
SEQ ID NO: 100857-101422

human_CDK9
SEQ ID NO: 101423-101531

human_CDX1
SEQ ID NO: 101532-101608

human_CDX2
SEQ ID NO: 101609-101700

human_CDX4
SEQ ID NO: 101701-101783

human_CEBPA
SEQ ID NO: 101784-101888

human_CEBPB
SEQ ID NO: 101889-101989

human_CEBPD
SEQ ID NO: 101990-102068

human_CEBPE
SEQ ID NO: 102069-102150

human_CEBPG
SEQ ID NO: 102151-102193

human_CEBPZ
SEQ ID NO: 102194-102507

human_CGGBP1
SEQ ID NO: 102508-102555

human_CHD2
SEQ ID NO: 102556-102703

human_CHD4
SEQ ID NO: 102704-103283

human_CHD7
SEQ ID NO: 103284-104180

human_CIART
SEQ ID NO: 104181-104293

human_CIITA
SEQ ID NO: 104294-104455

human_CITED1
SEQ ID NO: 104456-104511

human_CLOCK
SEQ ID NO: 104512-104763

human_CNBP
SEQ ID NO: 104764-104814

human_CREB1
SEQ ID NO: 104815-104910

human_CREB3
SEQ ID NO: 104911-105019

human_CREB3L1
SEQ ID NO: 105020-105173

human_CREB3L2
SEQ ID NO: 105174-105327

human_CREB3L3
SEQ ID NO: 105328-105463

human_CREB3L4
SEQ ID NO: 105464-105579

human_CREBBP
SEQ ID NO: 105580-106309

human_CREBRF
SEQ ID NO: 106310-106499

human_CREM
SEQ ID NO: 106500-106587

human_CRX
SEQ ID NO: 106588-106675

human_CRY1
SEQ ID NO: 106676-106849

human_CRY2
SEQ ID NO: 106850-107006

human_CT476828.9
SEQ ID NO: 107007-107131

human_CTCF
SEQ ID NO: 107132-107347

human_CTCFL
SEQ ID NO: 107348-107544

human_CUX1
SEQ ID NO: 107545-107993

human_CUX2
SEQ ID NO: 107994-108437

human_CXXC1
SEQ ID NO: 108438-108633

human_DACH1
SEQ ID NO: 108634-108843

human_DBP
SEQ ID NO: 108844-108938

human_DDIT3
SEQ ID NO: 108939-108987

human_DDN
SEQ ID NO: 108988-109198

human_DEAF1
SEQ ID NO: 109199-109365

human_DHX36
SEQ ID NO: 109366-109657

human_DHX9
SEQ ID NO: 109658-110036

human_DLX1
SEQ ID NO: 110037-110110

human_DLX2
SEQ ID NO: 110111-110206

human_DLX4
SEQ ID NO: 110207-110276

human_DLX5
SEQ ID NO: 110277-110361

human_DMBX1
SEQ ID NO: 110362-110472

human_DMRT1
SEQ ID NO: 110473-110582

human_DMRT2
SEQ ID NO: 110583-110748

human_DNMT3A
SEQ ID NO: 110749-111019

human_DPF2
SEQ ID NO: 111020-111079

human_DR1
SEQ ID NO: 111080-111130

human_DRAP1
SEQ ID NO: 111131-111189

human_DUX4
SEQ ID NO: 111190-111314

human_E2F1
SEQ ID NO: 111315-111443

human_E2F2
SEQ ID NO: 111444-111572

human_E2F3
SEQ ID NO: 111573-111709

human_E2F4
SEQ ID NO: 111710-111831

human_E2F6
SEQ ID NO: 111832-111904

human_E2F7
SEQ ID NO: 111905-112175

human_E2F8
SEQ ID NO: 112176-112433

human_E4F1
SEQ ID NO: 112434-112666

human_EAF2
SEQ ID NO: 112667-112742

human_EBF2
SEQ ID NO: 112743-112912

human_EBF3
SEQ ID NO: 112913-113075

human_EBF4
SEQ ID NO: 113076-113252

human_EED
SEQ ID NO: 113253-113390

human_EGR1
SEQ ID NO: 113391-113551

human_EGR2
SEQ ID NO: 113552-113692

human_EGR3
SEQ ID NO: 113693-113806

human_EGR4
SEQ ID NO: 113807-113981

human_EHF
SEQ ID NO: 113982-114069

human_EHMT2
SEQ ID NO: 114070-114437

human_ELF1
SEQ ID NO: 114438-114621

human_ELF3
SEQ ID NO: 114622-114730

human_ELF4
SEQ ID NO: 114731-114927

human_ELF5
SEQ ID NO: 114928-115004

human_ELK1
SEQ ID NO: 115005-115130

human_ELK3
SEQ ID NO: 115131-115250

human_ELK4
SEQ ID NO: 115251-115369

human_ELL3
SEQ ID NO: 115370-115486

human_ELMSAN1
SEQ ID NO: 115487-115797

human_EN1
SEQ ID NO: 115798-115912

human_ENO1
SEQ ID NO: 115913-116040

human_EOMES
SEQ ID NO: 116041-116249

human_EP300
SEQ ID NO: 116250-116971

human_ERBB4
SEQ ID NO: 116972-117361

human_ERG
SEQ ID NO: 117362-117503

human_ESR1
SEQ ID NO: 117504-117679

human_ESR2
SEQ ID NO: 117680-117809

human_ESRRA
SEQ ID NO: 117810-117934

human_ESRRB
SEQ ID NO: 117935-118084

human_ESRRG
SEQ ID NO: 118085-118223

human_ESX1
SEQ ID NO: 118224-118343

human_ETS1
SEQ ID NO: 118344-118473

human_ETS2
SEQ ID NO: 118474-118612

human_ETV1
SEQ ID NO: 118613-118748

human_ETV2
SEQ ID NO: 118749-118848

human_ETV3
SEQ ID NO: 118849-118999

human_ETV4
SEQ ID NO: 119000-119142

human_ETV5
SEQ ID NO: 119143-119293

human_ETV6
SEQ ID NO: 119294-119426

human_ETV7
SEQ ID NO: 119427-119526

human_EZH2
SEQ ID NO: 119527-119749

human_FERD3L
SEQ ID NO: 119750-119797

human_FEZF1
SEQ ID NO: 119798-119937

human_FEZF2
SEQ ID NO: 119938-120073

human_FIGLA
SEQ ID NO: 120074-120137

human_FLI1
SEQ ID NO: 120138-120270

human_FOS
SEQ ID NO: 120271-120382

human_FOSB
SEQ ID NO: 120383-120481

human_FOSL1
SEQ ID NO: 120482-120560

human_FOSL2
SEQ ID NO: 120561-120656

human_FOXA1
SEQ ID NO: 120657-120795

human_FOXA2
SEQ ID NO: 120796-120930

human_FOXA3
SEQ ID NO: 120931-121033

human_FOXC1
SEQ ID NO: 121034-121197

human_FOXC2
SEQ ID NO: 121198-121345

human_FOXD1
SEQ ID NO: 121346-121482

human_FOXD3
SEQ ID NO: 121483-121623

human_FOXF1
SEQ ID NO: 121624-121735

human_FOXF2
SEQ ID NO: 121736-121866

human_FOXH1
SEQ ID NO: 121867-121973

human_FOXI1
SEQ ID NO: 121974-122084

human_FOXJ1
SEQ ID NO: 122085-122208

human_FOXJ2
SEQ ID NO: 122209-122378

human_FOXK1
SEQ ID NO: 122379-122596

human_FOXK2
SEQ ID NO: 122597-122792

human_FOXL2
SEQ ID NO: 122793-122903

human_FOXM1
SEQ ID NO: 122904-123141

human_FOXN4
SEQ ID NO: 123142-123294

human_FOXO3
SEQ ID NO: 123295-123494

human_FOXP2
SEQ ID NO: 123495-123714

human_FOXP3
SEQ ID NO: 123715-123841

human_FOXQ1
SEQ ID NO: 123842-123960

human_FOXS1
SEQ ID NO: 123961-124057

human_FUBP3
SEQ ID NO: 124058-124226

human_GABPA
SEQ ID NO: 124227-124360

human_GABPB1
SEQ ID NO: 124361-124476

human_GABPB2
SEQ ID NO: 124477-124608

human_GADD45A
SEQ ID NO: 124609-124655

human_GATA1
SEQ ID NO: 124656-124777

human_GATA2
SEQ ID NO: 124778-124919

human_GATA3
SEQ ID NO: 124920-125050

human_GATA4
SEQ ID NO: 125051-125180

human_GATA5
SEQ ID NO: 125181-125297

human_GATA6
SEQ ID NO: 125298-125473

human_GATAD2B
SEQ ID NO: 125474-125649

human_GBX2
SEQ ID NO: 125650-125751

human_GCFC2
SEQ ID NO: 125752-125983

human_GCM1
SEQ ID NO: 125984-126112

human_GFI1
SEQ ID NO: 126113-126236

human_GLI1
SEQ ID NO: 126237-126566

human_GLI2
SEQ ID NO: 126567-127040

human_GLI3
SEQ ID NO: 127041-127512

human_GLIS1
SEQ ID NO: 127513-127696

human_GLIS2
SEQ ID NO: 127697-127851

human_GLMP
SEQ ID NO: 127852-127971

human_GMEB1
SEQ ID NO: 127972-128141

human_GMEB2
SEQ ID NO: 128142-128298

human_GRHL1
SEQ ID NO: 128299-128481

human_GRHL2
SEQ ID NO: 128482-128666

human_GSC
SEQ ID NO: 128667-128741

human_GSX1
SEQ ID NO: 128742-128818

human_GTF2B
SEQ ID NO: 128819-128911

human_GTF3C1
SEQ ID NO: 128912-129542

human_GZF1
SEQ ID NO: 129543-129753

human_H2AFY
SEQ ID NO: 129754-129862

human_H2AFY2
SEQ ID NO: 129863-129971

human_H2AFZ
SEQ ID NO: 129972-130007

human_H3F3A
SEQ ID NO: 130008-130046

human_H3F3B
SEQ ID NO: 130047-130085

human_HAND1
SEQ ID NO: 130086-130147

human_HAND2
SEQ ID NO: 130148-130210

human_HDAC1
SEQ ID NO: 130211-130352

human_HDAC2
SEQ ID NO: 130353-130487

human_HDAC4
SEQ ID NO: 130488-130810

human_HDAC5
SEQ ID NO: 130811-131145

human_HDAC6
SEQ ID NO: 131146-131507

human_HELT
SEQ ID NO: 131508-131577

human_HES1
SEQ ID NO: 131578-131659

human_HES2
SEQ ID NO: 131660-131709

human_HES3
SEQ ID NO: 131710-131763

human_HES4
SEQ ID NO: 131764-131827

human_HES5
SEQ ID NO: 131828-131875

human_HES6
SEQ ID NO: 131876-131940

human_HES7
SEQ ID NO: 131941-132005

human_HESX1
SEQ ID NO: 132006-132058

human_HEY1
SEQ ID NO: 132059-132148

human_HEY2
SEQ ID NO: 132149-132247

human_HEYL
SEQ ID NO: 132248-132343

human_HHEX
SEQ ID NO: 132344-132422

human_HIC2
SEQ ID NO: 132423-132604

human_HIF1A
SEQ ID NO: 132605-132857

human_HINFP
SEQ ID NO: 132858-133010

human_HIVEP1
SEQ ID NO: 133011-133823

human_HLF
SEQ ID NO: 133824-133909

human_HLTF
SEQ ID NO: 133910-134210

human_HMGA1
SEQ ID NO: 134211-134237

human_HMGA2
SEQ ID NO: 134238-134267

human_HMGB1
SEQ ID NO: 134268-134329

human_HMGB2
SEQ ID NO: 134330-134390

human_HMX1
SEQ ID NO: 134391-134492

human_HMX3
SEQ ID NO: 134493-134597

human_HNF1A
SEQ ID NO: 134598-134678

human_HNF1B
SEQ ID NO: 134679-134843

human_HNF4A
SEQ ID NO: 134844-134976

human_HNF4G
SEQ ID NO: 134977-135096

human_HNRNPC
SEQ ID NO: 135097-135186

human_HNRNPK
SEQ ID NO: 135187-135323

human_HNRNPL
SEQ ID NO: 135324-135498

human_HNRNPU
SEQ ID NO: 135499-135743

human_HOXA10
SEQ ID NO: 135744-135864

human_HOXA2
SEQ ID NO: 135865-135975

human_HOXA3
SEQ ID NO: 135976-136106

human_HOXA4
SEQ ID NO: 136107-136200

human_HOXA5
SEQ ID NO: 136201-136279

human_HOXA6
SEQ ID NO: 136280-136347

human_HOXA7
SEQ ID NO: 136348-136414

human_HOXA9
SEQ ID NO: 136415-136493

human_HOXB1
SEQ ID NO: 136494-136581

human_HOXB2
SEQ ID NO: 136582-136686

human_HOXB3
SEQ ID NO: 136687-136813

human_HOXB4
SEQ ID NO: 136814-136886

human_HOXB5
SEQ ID NO: 136887-136965

human_HOXB6
SEQ ID NO: 136966-137030

human_HOXB7
SEQ ID NO: 137031-137093

human_HOXB9
SEQ ID NO: 137094-137166

human_HOXC10
SEQ ID NO: 137167-137266

human_HOXC11
SEQ ID NO: 137267-137355

human_HOXC4
SEQ ID NO: 137356-137432

human_HOXC5
SEQ ID NO: 137433-137496

human_HOXC6
SEQ ID NO: 137497-137564

human_HOXD10
SEQ ID NO: 137565-137664

human_HOXD13
SEQ ID NO: 137665-137765

human_HOXD3
SEQ ID NO: 137766-137892

human_HOXD4
SEQ ID NO: 137893-137966

human_HOXD8
SEQ ID NO: 137967-138051

human_HOXD9
SEQ ID NO: 138052-138154

human_HR
SEQ ID NO: 138155-138492

human_HSF1
SEQ ID NO: 138493-138649

human_HSF2
SEQ ID NO: 138650-138808

human_HSF4
SEQ ID NO: 138809-138944

human_HSF5
SEQ ID NO: 138945-139121

human_HSFX1
SEQ ID NO: 139122-139246

human_HSFX2
SEQ ID NO: 139247-139371

human_HSFX3
SEQ ID NO: 139372-139469

human_HSFX4
SEQ ID NO: 139470-139567

human_HSFY1
SEQ ID NO: 139568-139626

human_HSFY2
SEQ ID NO: 139627-139744

human_IER2
SEQ ID NO: 139745-139809

human_IFI16
SEQ ID NO: 139810-140042

human_IKZF1
SEQ ID NO: 140043-140196

human_IKZF2
SEQ ID NO: 140197-140209

human_IKZF3
SEQ ID NO: 140210-140286

human_IKZF4
SEQ ID NO: 140287-140459

human_IKZF5
SEQ ID NO: 140460-140583

human_INSM1
SEQ ID NO: 140584-140734

human_IRF1
SEQ ID NO: 140735-140829

human_IRF2
SEQ ID NO: 140830-140932

human_IRF2BP1
SEQ ID NO: 140933-141105

human_IRF2BP2
SEQ ID NO: 141106-141279

human_IRF2BPL
SEQ ID NO: 141280-141516

human_IRF3
SEQ ID NO: 141517-141642

human_IRF4
SEQ ID NO: 141643-141775

human_IRF5
SEQ ID NO: 141776-141922

human_IRF6
SEQ ID NO: 141923-142060

human_IRF7
SEQ ID NO: 142061-142213

human_IRF8
SEQ ID NO: 142214-142339

human_IRF9
SEQ ID NO: 142340-142400

human_ISL1
SEQ ID NO: 142401-142503

human_JARID2
SEQ ID NO: 142504-142875

human_JDP2
SEQ ID NO: 142876-142925

human_JMJD1C
SEQ ID NO: 142926-143685

human_JUN
SEQ ID NO: 143686-143782

human_JUNB
SEQ ID NO: 143783-143884

human_JUND
SEQ ID NO: 143885-143986

human_KAT2B
SEQ ID NO: 143987-144233

human_KAT7
SEQ ID NO: 144234-144414

human_KCNIP3
SEQ ID NO: 144415-144489

human_KDM1A
SEQ ID NO: 144490-144750

human_KDM2B
SEQ ID NO: 144751-145149

human_KDM3A
SEQ ID NO: 145150-145543

human_KDM3B
SEQ ID NO: 145544-146069

human_KDM5A
SEQ ID NO: 146070-146574

human_KDM6A
SEQ ID NO: 146575-146992

human_KDM6B
SEQ ID NO: 146993-147494

human_KLF1
SEQ ID NO: 147495-147600

human_KLF10
SEQ ID NO: 147601-147742

human_KLF11
SEQ ID NO: 147743-147893

human_KLF12
SEQ ID NO: 147894-148011

human_KLF13
SEQ ID NO: 148012-148095

human_KLF15
SEQ ID NO: 148096-148218

human_KLF16
SEQ ID NO: 148219-148291

human_KLF17
SEQ ID NO: 148292-148406

human_KLF3
SEQ ID NO: 148407-148507

human_KLF4
SEQ ID NO: 148508-148649

human_KLF5
SEQ ID NO: 148650-148784

human_KLF6
SEQ ID NO: 148785-148866

human_KLF7
SEQ ID NO: 148867-148954

human_KLF8
SEQ ID NO: 148955-149029

human_KMT2A
SEQ ID NO: 149030-150218

human_KMT2D
SEQ ID NO: 150219-151877

human_LDB1
SEQ ID NO: 151878-151998

human_LEF1
SEQ ID NO: 151999-152116

human_LHX2
SEQ ID NO: 152117-152236

human_LHX3
SEQ ID NO: 152237-152354

human_LITAF
SEQ ID NO: 152355-152400

human_LMO2
SEQ ID NO: 152401-152466

human_LMO4
SEQ ID NO: 152467-152513

human_LMX1A
SEQ ID NO: 152514-152625

human_LMX1B
SEQ ID NO: 152626-152745

human_LONP1
SEQ ID NO: 152746-153031

human_LRRFIP1
SEQ ID NO: 153032-153264

human_LYL1
SEQ ID NO: 153265-153346

human_MACC1
SEQ ID NO: 153347-153599

human_MAF
SEQ ID NO: 153600-153718

human_MAF1
SEQ ID NO: 153719-153793

human_MAFA
SEQ ID NO: 153794-153897

human_MAFB
SEQ ID NO: 153898-153992

human_MAFF
SEQ ID NO: 153993-154039

human_MAFG
SEQ ID NO: 154040-154085

human_MAFK
SEQ ID NO: 154086-154130

human_MAX
SEQ ID NO: 154131-154157

human_MAZ
SEQ ID NO: 154158-154303

human_MBD2
SEQ ID NO: 154304-154424

human_MBD3
SEQ ID NO: 154425-154500

human_MED1
SEQ ID NO: 154501-154972

human_MED12
SEQ ID NO: 154973-155623

human_MED8
SEQ ID NO: 155624-155711

human_MEF2A
SEQ ID NO: 155712-155858

human_MEF2B
SEQ ID NO: 155859-155968

human_MEF2C
SEQ ID NO: 155969-156111

human_MEF2D
SEQ ID NO: 156112-156265

human_MEIS1
SEQ ID NO: 156266-156380

human_MEIS2
SEQ ID NO: 156381-156468

human_MEN1
SEQ ID NO: 156469-156650

human_MEOX1
SEQ ID NO: 156651-156724

human_MEOX2
SEQ ID NO: 156725-156813

human_MESP1
SEQ ID NO: 156814-156891

human_MESP2
SEQ ID NO: 156892-157008

human_MITF
SEQ ID NO: 157009-157162

human_MIXL1
SEQ ID NO: 157163-157229

human_MLX
SEQ ID NO: 157230-157316

human_MLXIP
SEQ ID NO: 157317-157590

human_MLXIPL
SEQ ID NO: 157591-157760

human_MMP12
SEQ ID NO: 157761-157899

human_MNT
SEQ ID NO: 157900-158071

human_MRTFA
SEQ ID NO: 158072-158359

human_MSC
SEQ ID NO: 158360-158419

human_MSGN1
SEQ ID NO: 158420-158475

human_MSX1
SEQ ID NO: 158476-158564

human_MSX2
SEQ ID NO: 158565-158642

human_MTA1
SEQ ID NO: 158643-158854

human_MTA2
SEQ ID NO: 158855-159052

human_MTERF3
SEQ ID NO: 159053-159175

SEQ ID NOs: 159176-244000

human_MTF1
SEQ ID NO: 159176-159399

human_MTF2
SEQ ID NO: 159400-159575

human_MTOR
SEQ ID NO: 159576-160338

human_MUC1
SEQ ID NO: 160339-160418

human_MXD1
SEQ ID NO: 160419-160482

human_MXD3
SEQ ID NO: 160483-160542

human_MXI1
SEQ ID NO: 160543-160608

human_MYB
SEQ ID NO: 160609-160834

human_MYBBP1A
SEQ ID NO: 160835-161231

human_MYBL1
SEQ ID NO: 161232-161454

human_MYBL2
SEQ ID NO: 161455-161662

human_MYC
SEQ ID NO: 161663-161792

human_MYCN
SEQ ID NO: 161793-161929

human_MYEF2
SEQ ID NO: 161930-161965

human_MYF5
SEQ ID NO: 161966-162039

human_MYF6
SEQ ID NO: 162040-162109

human_MYOCD
SEQ ID NO: 162110-162388

human_MYOD1
SEQ ID NO: 162389-162482

human_MYOG
SEQ ID NO: 162483-162547

human_MYPOP
SEQ ID NO: 162548-162665

human_MYT1
SEQ ID NO: 162666-162999

human_MYT1L
SEQ ID NO: 163000-163352

human_MZF1
SEQ ID NO: 163353-163570

human_NACC2
SEQ ID NO: 163571-163744

human_NANOG
SEQ ID NO: 163745-163833

human_NCOA2
SEQ ID NO: 163834-164270

human_NCOR1
SEQ ID NO: 164271-165000

human_NCOR2
SEQ ID NO: 165001-165752

human_NDN
SEQ ID NO: 165753-165846

human_NEUROD1
SEQ ID NO: 165847-165951

human_NEUROD2
SEQ ID NO: 165952-166063

human_NEUROD6
SEQ ID NO: 166064-166162

human_NEUROG1
SEQ ID NO: 166163-166231

human_NEUROG2
SEQ ID NO: 166232-166310

human_NEUROG3
SEQ ID NO: 166311-166372

human_NFAT5
SEQ ID NO: 166373-166829

human_NFATC1
SEQ ID NO: 166830-167074

human_NFATC2
SEQ ID NO: 167075-167348

human_NFATC3
SEQ ID NO: 167349-167668

human_NFATC4
SEQ ID NO: 167669-167936

human_NFE2
SEQ ID NO: 167937-168046

human_NFE2L1
SEQ ID NO: 168047-168266

human_NFE2L2
SEQ ID NO: 168267-168445

human_NFE2L3
SEQ ID NO: 168446-168651

human_NFIA
SEQ ID NO: 168652-168799

human_NFIB
SEQ ID NO: 168800-168923

human_NFIC
SEQ ID NO: 168924-169049

human_NFIL3
SEQ ID NO: 169050-169185

human_NFKB1
SEQ ID NO: 169186-169474

human_NFKB2
SEQ ID NO: 169475-169742

human_NFX1
SEQ ID NO: 169743-169990

human_NFXL1
SEQ ID NO: 169991-170261

human_NFYA
SEQ ID NO: 170262-170354

human_NFYB
SEQ ID NO: 170355-170414

human_NFYC
SEQ ID NO: 170415-170549

human_NHLH1
SEQ ID NO: 170550-170587

human_NHLH2
SEQ ID NO: 170588-170625

human_NKRF
SEQ ID NO: 170626-170830

human_NKX2-1
SEQ ID NO: 170831-170948

human_NKX2-2
SEQ ID NO: 170949-171028

human_NKX2-5
SEQ ID NO: 171029-171123

human_NKX2-6
SEQ ID NO: 171124-171211

human_NKX2-8
SEQ ID NO: 171212-171281

human_NKX3-1
SEQ ID NO: 171282-171349

human_NKX3-2
SEQ ID NO: 171350-171447

human_NKX6-1
SEQ ID NO: 171448-171555

human_NKX6-2
SEQ ID NO: 171556-171636

human_NLRC5
SEQ ID NO: 171637-172194

human_NME1
SEQ ID NO: 172195-172245

human_NONO
SEQ ID NO: 172246-172384

human_NOTCH1
SEQ ID NO: 172385-173148

human_NPAS2
SEQ ID NO: 173149-173393

human_NPAS4
SEQ ID NO: 173394-173631

human_NPM1
SEQ ID NO: 173632-173717

human_NR1D1
SEQ ID NO: 173718-173899

human_NR1D2
SEQ ID NO: 173900-174071

human_NR1H2
SEQ ID NO: 174072-174207

human_NR1H3
SEQ ID NO: 174208-174321

human_NR1H4
SEQ ID NO: 174322-174463

human_NR1I2
SEQ ID NO: 174464-174603

human_NR1I3
SEQ ID NO: 174604-174706

human_NR2C1
SEQ ID NO: 174707-174885

human_NR2C2
SEQ ID NO: 174886-175067

human_NR2E3
SEQ ID NO: 175068-175188

human_NR2F1
SEQ ID NO: 175189-175313

human_NR2F6
SEQ ID NO: 175314-175432

human_NR3C1
SEQ ID NO: 175433-175663

human_NR4A1
SEQ ID NO: 175664-175840

human_NR4A2
SEQ ID NO: 175841-176017

human_NR4A3
SEQ ID NO: 176018-176206

human_NR5A1
SEQ ID NO: 176207-176342

human_NR5A2
SEQ ID NO: 176343-176488

human_NR6A1
SEQ ID NO: 176489-176630

human_NRF1
SEQ ID NO: 176631-176779

human_NRIP1
SEQ ID NO: 176780-177124

human_NRL
SEQ ID NO: 177125-177193

human_NSD1
SEQ ID NO: 177194-177919

human_ONECUT2
SEQ ID NO: 177920-178068

human_ONECUT3
SEQ ID NO: 178069-178214

human_OSR1
SEQ ID NO: 178215-178292

human_OSR2
SEQ ID NO: 178293-178383

human_OTX1
SEQ ID NO: 178384-178487

human_OTX2
SEQ ID NO: 178488-178574

human_OVOL1
SEQ ID NO: 178575-178652

human_PARP1
SEQ ID NO: 178653-178954

human_PATZ1
SEQ ID NO: 178955-179158

human_PAX1
SEQ ID NO: 179159-179316

human_PAX2
SEQ ID NO: 179317-179432

human_PAX4
SEQ ID NO: 179433-179533

human_PAX5
SEQ ID NO: 179534-179648

human_PAX6
SEQ ID NO: 179649-179772

human_PAX8
SEQ ID NO: 179773-179866

human_PAX9
SEQ ID NO: 179867-179966

human_PAXBP1
SEQ ID NO: 179967-180208

human_PBX1
SEQ ID NO: 180209-180335

human_PBX2
SEQ ID NO: 180336-180462

human_PBX3
SEQ ID NO: 180463-180590

human_PCGF3
SEQ ID NO: 180591-180660

human_PCGF5
SEQ ID NO: 180661-180735

human_PCGF6
SEQ ID NO: 180736-180815

human_PDX1
SEQ ID NO: 180816-180898

human_PER1
SEQ ID NO: 180899-181283

human_PER2
SEQ ID NO: 181284-181657

human_PER3
SEQ ID NO: 181658-182018

human_PGR
SEQ ID NO: 182019-182265

human_PHB
SEQ ID NO: 182266-182344

human_PHOX2A
SEQ ID NO: 182345-182427

human_PHOX2B
SEQ ID NO: 182428-182519

human_PIH1D1
SEQ ID NO: 182520-182604

human_PITX1
SEQ ID NO: 182605-182696

human_PITX2
SEQ ID NO: 182697-182791

human_PITX3
SEQ ID NO: 182792-182879

human_PKNOX2
SEQ ID NO: 182880-183018

human_PLAG1
SEQ ID NO: 183019-183166

human_PLAGL1
SEQ ID NO: 183167-183303

human_POLRMT
SEQ ID NO: 183304-183670

human_POU1F1
SEQ ID NO: 183671-183763

human_POU2AF1
SEQ ID NO: 183764-183838

human_POU2F1
SEQ ID NO: 183839-184062

human_POU2F2
SEQ ID NO: 184063-184204

human_POU2F3
SEQ ID NO: 184205-184333

human_POU3F2
SEQ ID NO: 184334-184464

human_POU3F4
SEQ ID NO: 184465-184570

human_POU4F1
SEQ ID NO: 184571-184694

human_POU4F2
SEQ ID NO: 184695-184815

human_POU4F3
SEQ ID NO: 184816-184914

human_POU5F1
SEQ ID NO: 184915-185020

human_POU6F1
SEQ ID NO: 185021-185201

human_PPARA
SEQ ID NO: 185202-185339

human_PPARD
SEQ ID NO: 185340-185445

human_PPARG
SEQ ID NO: 185446-185594

human_PRDM1
SEQ ID NO: 185595-185829

human_PRDM11
SEQ ID NO: 185830-185970

human_PRDM12
SEQ ID NO: 185971-186078

human_PRDM13
SEQ ID NO: 186079-186288

human_PRDM14
SEQ ID NO: 186289-186457

human_PRDM15
SEQ ID NO: 186458-186907

human_PRDM2
SEQ ID NO: 186908-187420

human_PRDM4
SEQ ID NO: 187421-187658

human_PRDM5
SEQ ID NO: 187659-187845

human_PRDM6
SEQ ID NO: 187846-188021

human_PRDM7
SEQ ID NO: 188022-188166

human_PRDM9
SEQ ID NO: 188167-188432

human_PRDX5
SEQ ID NO: 188433-188494

human_PRKN
SEQ ID NO: 188495-188574

human_PRMT5
SEQ ID NO: 188575-188745

human_PROP1
SEQ ID NO: 188746-188811

human_PROX1
SEQ ID NO: 188812-189030

human_PRRX1
SEQ ID NO: 189031-189101

human_PSPC1
SEQ ID NO: 189102-189256

human_PTF1A
SEQ ID NO: 189257-189352

human_PURA
SEQ ID NO: 189353-189446

human_PURB
SEQ ID NO: 189447-189537

human_PURG
SEQ ID NO: 189538-189631

human_RAI1
SEQ ID NO: 189632-190201

human_RARA
SEQ ID NO: 190202-190337

human_RARB
SEQ ID NO: 190338-190469

human_RARG
SEQ ID NO: 190470-190600

human_RAX
SEQ ID NO: 190601-190629

human_RAX2
SEQ ID NO: 190630-190682

human_RB1
SEQ ID NO: 190683-190958

human_RBBP4
SEQ ID NO: 190959-191083

human_RBBP5
SEQ ID NO: 191084-191242

human_RBL1
SEQ ID NO: 191243-191544

human_RBL2
SEQ ID NO: 191545-191884

human_RBMX
SEQ ID NO: 191885-191999

human_RBPJ
SEQ ID NO: 192000-192147

human_RBPJL
SEQ ID NO: 192148-192300

human_RCOR1
SEQ ID NO: 192301-192443

human_RCOR2
SEQ ID NO: 192444-192598

human_RCOR3
SEQ ID NO: 192599-192744

human_REL
SEQ ID NO: 192745-192928

human_RELA
SEQ ID NO: 192929-193090

human_RELB
SEQ ID NO: 193091-193262

human_REST
SEQ ID NO: 193263-193589

human_RFX1
SEQ ID NO: 193590-193881

human_RFX2
SEQ ID NO: 193882-194096

human_RFX3
SEQ ID NO: 194097-194218

human_RFX4
SEQ ID NO: 194219-194408

human_RFX5
SEQ ID NO: 194409-194591

human_RFX6
SEQ ID NO: 194592-194867

human_RFX7
SEQ ID NO: 194868-195274

human_RFX8
SEQ ID NO: 195275-195414

human_RNF10
SEQ ID NO: 195415-195655

human_RORA
SEQ ID NO: 195656-195820

human_RORB
SEQ ID NO: 195821-195956

human_RORC
SEQ ID NO: 195957-196109

human_RPS3
SEQ ID NO: 196110-196185

human_RPTOR
SEQ ID NO: 196186-196583

human_RREB1
SEQ ID NO: 196584-197024

human_RRN3
SEQ ID NO: 197025-197217

human_RUNX1
SEQ ID NO: 197218-197359

human_RUNX2
SEQ ID NO: 197360-197513

human_RUNX3
SEQ ID NO: 197514-197640

human_RUVBL2
SEQ ID NO: 197641-197777

human_RXRA
SEQ ID NO: 197778-197913

human_RXRB
SEQ ID NO: 197914-198071

human_SAFB
SEQ ID NO: 198072-198343

human_SALL1
SEQ ID NO: 198344-198738

human_SALL2
SEQ ID NO: 198739-199038

human_SARS
SEQ ID NO: 199039-199190

human_SATB1
SEQ ID NO: 199191-199417

human_SATB2
SEQ ID NO: 199418-199635

human_SCRT1
SEQ ID NO: 199636-199737

human_SCRT2
SEQ ID NO: 199738-199827

human_SCX
SEQ ID NO: 199828-199885

human_SETX
SEQ ID NO: 199886-200686

human_SFPQ
SEQ ID NO: 200687-200896

human_SIN3A
SEQ ID NO: 200897-201276

human_SIRT1
SEQ ID NO: 201277-201498

human_SIX1
SEQ ID NO: 201499-201581

human_SIX2
SEQ ID NO: 201582-201666

human_SIX3
SEQ ID NO: 201667-201763

human_SIX4
SEQ ID NO: 201764-201995

human_SIX5
SEQ ID NO: 201996-202215

human_SIX6
SEQ ID NO: 202216-202287

human_SKIL
SEQ ID NO: 202288-202490

human_SMAD1
SEQ ID NO: 202491-202627

human_SMAD2
SEQ ID NO: 202628-202765

human_SMAD3
SEQ ID NO: 202766-202890

human_SMAD4
SEQ ID NO: 202891-203053

human_SMAD5
SEQ ID NO: 203054-203190

human_SMAD6
SEQ ID NO: 203191-203337

human_SMAD7
SEQ ID NO: 203338-203463

human_SMARCA2
SEQ ID NO: 203464-203938

human_SMARCA4
SEQ ID NO: 203939-204430

human_SMARCB1
SEQ ID NO: 204431-204541

human_SMARCC1
SEQ ID NO: 204542-204870

human_SMARCC2
SEQ ID NO: 204871-205232

human_SMARCD2
SEQ ID NO: 205233-205375

human_SMARCE1
SEQ ID NO: 205376-205482

human_SMYD3
SEQ ID NO: 205483-205608

human_SNAI1
SEQ ID NO: 205609-205685

human_SNAI2
SEQ ID NO: 205686-205763

human_SNAI3
SEQ ID NO: 205764-205848

human_SNCA
SEQ ID NO: 205849-205879

human_SOX1
SEQ ID NO: 205880-205994

human_SOX10
SEQ ID NO: 205995-206132

human_SOX11
SEQ ID NO: 206133-206262

human_SOX12
SEQ ID NO: 206263-206354

human_SOX13
SEQ ID NO: 206355-206538

human_SOX17
SEQ ID NO: 206539-206660

human_SOX18
SEQ ID NO: 206661-206773

human_SOX2
SEQ ID NO: 206774-206866

human_SOX21
SEQ ID NO: 206867-206947

human_SOX3
SEQ ID NO: 206948-207079

human_SOX4
SEQ ID NO: 207080-207219

human_SOX6
SEQ ID NO: 207220-207459

human_SOX7
SEQ ID NO: 207460-207573

human_SOX8
SEQ ID NO: 207574-207705

human_SOX9
SEQ ID NO: 207706-207856

human_SP1
SEQ ID NO: 207857-208089

human_SP2
SEQ ID NO: 208090-208271

human_SP3
SEQ ID NO: 208272-208503

human_SP5
SEQ ID NO: 208504-208620

human_SP7
SEQ ID NO: 208621-208747

human_SPI1
SEQ ID NO: 208748-208826

human_SPIB
SEQ ID NO: 208827-208877

human_SPIC
SEQ ID NO: 208878-208949

human_SREBF1
SEQ ID NO: 208950-209291

human_SREBF2
SEQ ID NO: 209292-209631

human_SRF
SEQ ID NO: 209632-209781

human_SSBP2
SEQ ID NO: 209782-209887

human_SSBP3
SEQ ID NO: 209888-209995

human_SSBP4
SEQ ID NO: 209996-210108

human_ST18
SEQ ID NO: 210109-210420

human_STAT1
SEQ ID NO: 210421-210631

human_STAT3
SEQ ID NO: 210632-210860

human_STAT5B
SEQ ID NO: 210861-211094

human_STAT6
SEQ ID NO: 211095-211346

human_STOX1
SEQ ID NO: 211347-211641

human_SUV39H1
SEQ ID NO: 211642-211766

human_SUV39H2
SEQ ID NO: 211767-211887

human_SUZ12
SEQ ID NO: 211888-212107

human_TAF1
SEQ ID NO: 212108-212673

human_TAF1B
SEQ ID NO: 212674-212847

human_TAF1C
SEQ ID NO: 212848-213077

human_TAF2
SEQ ID NO: 213078-213435

human_TAF5
SEQ ID NO: 213436-213673

human_TAF7
SEQ ID NO: 213674-213776

human_TAF7L
SEQ ID NO: 213777-213864

human_TAF9
SEQ ID NO: 213865-213941

human_TAF9B
SEQ ID NO: 213942-214014

human_TAL1
SEQ ID NO: 214015-214111

human_TAL2
SEQ ID NO: 214112-214141

human_TBL1X
SEQ ID NO: 214142-214312

human_TBL1XR1
SEQ ID NO: 214313-214464

human_TBP
SEQ ID NO: 214465-214564

human_TBPL1
SEQ ID NO: 214565-214618

human_TBPL2
SEQ ID NO: 214619-214728

human_TBR1
SEQ ID NO: 214729-214930

human_TBX15
SEQ ID NO: 214931-215108

human_TBX18
SEQ ID NO: 215109-215288

human_TBX19
SEQ ID NO: 215289-215420

human_TBX2
SEQ ID NO: 215421-215631

human_TBX20
SEQ ID NO: 215632-215763

human_TBX21
SEQ ID NO: 215764-215921

human_TBX22
SEQ ID NO: 215922-216075

human_TBX3
SEQ ID NO: 216076-216296

human_TBX5
SEQ ID NO: 216297-216449

human_TBX6
SEQ ID NO: 216450-216578

human_TBXT
SEQ ID NO: 216579-216706

human_TCF12
SEQ ID NO: 216707-216908

human_TCF15
SEQ ID NO: 216909-216966

human_TCF20
SEQ ID NO: 216967-217545

human_TCF21
SEQ ID NO: 217546-217597

human_TCF3
SEQ ID NO: 217598-217791

human_TCF4
SEQ ID NO: 217792-217990

human_TCF7
SEQ ID NO: 217991-218103

human_TCF7L1
SEQ ID NO: 218104-218277

human_TCF7L2
SEQ ID NO: 218278-218413

human_TCFL5
SEQ ID NO: 218414-218561

human_TEAD1
SEQ ID NO: 218562-218666

human_TEAD2
SEQ ID NO: 218667-218798

human_TEAD3
SEQ ID NO: 218799-218926

human_TEAD4
SEQ ID NO: 218927-219015

human_TEF
SEQ ID NO: 219016-219104

human_TFAM
SEQ ID NO: 219105-219166

human_TFAP2A
SEQ ID NO: 219167-219294

human_TFAP2B
SEQ ID NO: 219295-219430

human_TFAP2C
SEQ ID NO: 219431-219563

human_TFAP2D
SEQ ID NO: 219564-219696

human_TFAP2E
SEQ ID NO: 219697-219826

human_TFAP4
SEQ ID NO: 219827-219925

human_TFCP2
SEQ ID NO: 219926-220073

human_TFCP2L1
SEQ ID NO: 220074-220215

human_TFDP1
SEQ ID NO: 220216-220336

human_TFDP2
SEQ ID NO: 220337-220449

human_TFE3
SEQ ID NO: 220450-220619

human_TFEB
SEQ ID NO: 220620-220760

human_TFEC
SEQ ID NO: 220761-220862

human_TGIF1
SEQ ID NO: 220863-220980

human_THAP1
SEQ ID NO: 220981-221042

human_THAP11
SEQ ID NO: 221043-221134

human_THRA
SEQ ID NO: 221135-221279

human_THRAP3
SEQ ID NO: 221280-221563

human_THRB
SEQ ID NO: 221564-221704

human_TIPARP
SEQ ID NO: 221705-221899

human_TLX1
SEQ ID NO: 221900-221996

human_TNF
SEQ ID NO: 221997-222064

human_TOP1
SEQ ID NO: 222065-222291

human_TOX2
SEQ ID NO: 222292-222441

human_TOX3
SEQ ID NO: 222442-222612

human_TP53
SEQ ID NO: 222613-222728

human_TP63
SEQ ID NO: 222729-222930

human_TP73
SEQ ID NO: 222931-223078

human_TRERF1
SEQ ID NO: 223079-223415

human_TRIM24
SEQ ID NO: 223416-223728

human_TRPS1
SEQ ID NO: 223729-224110

human_TWIST1
SEQ ID NO: 224111-224168

human_TXK
SEQ ID NO: 224169-224324

human_UBTF
SEQ ID NO: 224325-224551

human_UHRF1
SEQ ID NO: 224552-224787

human_USP3
SEQ ID NO: 224788-224941

human_UTY
SEQ ID NO: 224942-225263

human_VAX1
SEQ ID NO: 225264-225361

human_VAX2
SEQ ID NO: 225362-225446

human_VDR
SEQ ID NO: 225447-225572

human_VEZF1
SEQ ID NO: 225573-225726

human_WBP2
SEQ ID NO: 225727-225802

human_WNT1
SEQ ID NO: 225803-225911

human_WNT11
SEQ ID NO: 225912-226015

human_WNT5A
SEQ ID NO: 226016-226127

human_WT1
SEQ ID NO: 226128-226215

human_XBP1
SEQ ID NO: 226216-226291

human_XRCC5
SEQ ID NO: 226292-226508

human_XRCC6
SEQ ID NO: 226509-226689

human_XRN2
SEQ ID NO: 226690-226972

human_YAP1
SEQ ID NO: 226973-227121

human_YBX1
SEQ ID NO: 227122-227216

human_YBX3
SEQ ID NO: 227217-227325

human_YY1
SEQ ID NO: 227326-227447

human_YY2
SEQ ID NO: 227448-227556

human_ZBED1
SEQ ID NO: 227557-227762

human_ZBTB14
SEQ ID NO: 227763-227895

human_ZBTB16
SEQ ID NO: 227896-228095

human_ZBTB17
SEQ ID NO: 228096-228336

human_ZBTB2
SEQ ID NO: 228337-228488

human_ZBTB20
SEQ ID NO: 228489-228686

human_ZBTB24
SEQ ID NO: 228687-228893

human_ZBTB4
SEQ ID NO: 228894-229195

human_ZBTB48
SEQ ID NO: 229196-229399

human_ZBTB5
SEQ ID NO: 229400-229600

human_ZBTB7A
SEQ ID NO: 229601-229773

human_ZBTB7B
SEQ ID NO: 229774-229933

human_ZC3H4
SEQ ID NO: 229934-230322

human_ZC3H6
SEQ ID NO: 230323-230677

human_ZC3H8
SEQ ID NO: 230678-230762

human_ZEB1
SEQ ID NO: 230763-231097

human_ZFHX2
SEQ ID NO: 231098-231866

human_ZFHX3
SEQ ID NO: 231867-232975

human_ZFHX4
SEQ ID NO: 232976-234058

human_ZFP42
SEQ ID NO: 234059-234149

human_ZFPM1
SEQ ID NO: 234150-234449

human_ZGPAT
SEQ ID NO: 234450-234606

human_ZHX3
SEQ ID NO: 234607-234891

human_ZIC1
SEQ ID NO: 234892-235023

human_ZIC2
SEQ ID NO: 235024-235180

human_ZIC3
SEQ ID NO: 235181-235318

human_ZIC4
SEQ ID NO: 235319-235427

human_ZIC5
SEQ ID NO: 235428-235624

human_ZKSCAN3
SEQ ID NO: 235625-235783

human_ZNF131
SEQ ID NO: 235784-235958

human_ZNF143
SEQ ID NO: 235959-236138

human_ZNF148
SEQ ID NO: 236139-236374

human_ZNF174
SEQ ID NO: 236375-236494

human_ZNF175
SEQ ID NO: 236495-236705

human_ZNF202
SEQ ID NO: 236706-236897

human_ZNF205
SEQ ID NO: 236898-237061

human_ZNF217
SEQ ID NO: 237062-237373

human_ZNF219
SEQ ID NO: 237374-237587

human_ZNF239
SEQ ID NO: 237588-237722

human_ZNF277
SEQ ID NO: 237723-237855

human_ZNF281
SEQ ID NO: 237856-238121

human_ZNF322
SEQ ID NO: 238122-238239

human_ZNF335
SEQ ID NO: 238240-238639

human_ZNF350
SEQ ID NO: 238640-238796

human_ZNF395
SEQ ID NO: 238797-238948

human_ZNF431
SEQ ID NO: 238949-239119

human_ZNF497
SEQ ID NO: 239120-239266

human_ZNF501
SEQ ID NO: 239267-239345

human_ZNF513
SEQ ID NO: 239346-239505

human_ZNF516
SEQ ID NO: 239506-239852

human_ZNF536
SEQ ID NO: 239853-240240

human_ZNF541
SEQ ID NO: 240241-240642

human_ZNF564
SEQ ID NO: 240643-240806

human_ZNF568
SEQ ID NO: 240807-240997

human_ZNF589
SEQ ID NO: 240998-241104

human_ZNF605
SEQ ID NO: 241105-241303

human_ZNF613
SEQ ID NO: 241304-241486

human_ZNF639
SEQ ID NO: 241487-241629

human_ZNF649
SEQ ID NO: 241630-241778

human_ZNF658
SEQ ID NO: 241779-241961

human_ZNF668
SEQ ID NO: 241962-242151

human_ZNF691
SEQ ID NO: 242152-242243

human_ZNF692
SEQ ID NO: 242244-242397

human_ZNF704
SEQ ID NO: 242398-242518

human_ZNF709
SEQ ID NO: 242519-242708

human_ZNF711
SEQ ID NO: 242709-242934

human_ZNF740
SEQ ID NO: 242935-242990

human_ZNF746
SEQ ID NO: 242991-243181

human_ZNF750
SEQ ID NO: 243182-243396

human_ZNF821
SEQ ID NO: 243397-243517

human_ZNF835
SEQ ID NO: 243518-243676

human_ZNF93
SEQ ID NO: 243677-243860

human_ZSCAN21
SEQ ID NO: 243861-244000

REFERENCES

1. F. J. Isaacs, D. J. Dwyer, J. J. Collins, RNA synthetic biology. Nature biotechnology 24, 545 (2006).

2. A. A. Green, P. A. Silver, J. J. Collins, P. Yin, Toehold switches: de-novo-designed regulators of gene expression. Cell 159, 925-939 (2014).

3. K. Pardee et al., Rapid, low-cost detection of Zika virus using programmable biomolecular components. Cell 165, 1255-1266 (2016).

4. M. K. Takahashi et al., A low-cost paper-based synthetic biology platform for analyzing gut microbiota and host biomarkers. Nature communications 9, 3347 (2018).

5. A. A. Green et al., Complex cellular logic computation using ribocomputing devices. Nature 548, 117 (2017).

6. S.-J. Kim, M. Leong, M. B. Amrofell, Y. J. Lee, T. S. Moon, Modulating responses of toehold switches by an inhibitory hairpin. ACS synthetic biology 8, 601-605 (2019).

7. M. Krishnamurthy et al., Tunable riboregulator switches for post-transcriptional control of gene expression. ACS synthetic biology 4, 1326-1334 (2015).

8. J. Kim et al., De-Novo-Designed Translational Repressors for Multi-Input Cellular Logic. bioRxiv, 501783 (2018).

9. A. C.-Y. To et al., A comprehensive web tool for toehold switch design. Bioinformatics 34, 2862-2864 (2018).

10. H. K. Kim et al., Deep learning improves prediction of CRISPR-Cpf1 guide RNA activity. Nature biotechnology 36, 239 (2018).

11. S. Webb, Deep learning for biology. Nature 554, (2018).

12. C. Angermueller, T. Pärnamaa, L. Parts, O. Stegle, Deep learning for computational biology. Molecular systems biology 12, (2016).

13. M. Wainberg, D. Merico, A. Delong, B. J. Frey, Deep learning in biomedicine. Nature biotechnology 36, 829 (2018).

14. D. M. Camacho, K. M. Collins, R. K. Powers, J. C. Costello, J. J. Collins, Next-generation machine learning for biological networks. Cell 173, 1581-1592 (2018).

15. K. Pardee et al., Paper-based synthetic gene networks. Cell 159, 940-954 (2014).

16. D. Ma, L. Shen, K. Wu, C. W. Diehnelt, A. A. Green, Low-cost detection of norovirus using paper-based cell-free systems and synbody-based viral enrichment. Synthetic Biology 3, ysy018 (2018).

17. G. Chuai et al., DeepCRISPR: optimized CRISPR guide RNA design by deep learning. Genome biology 19, 80 (2018).

18. J. Luo, W. Chen, L. Xue, B. Tang, Prediction of activity and specificity of CRISPR-Cpf1 using convolutional deep learning neural networks. BMC bioinformatics 20, 332 (2019).

19. S. Zhang, H. Hu, T. Jiang, L. Zhang, J. Zeng, TITER: predicting translation initiation sites by deep learning. Bioinformatics 33, i234-i242 (2017).

20. J. Zuallaert, M. Kim, Y. Saeys, W. De Neve, in 2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). (IEEE, 2017), pp. 1233-1237.

21. E. C. Alley, G. Khimulya, S. Biswas, M. AlQuraishi, G. M. Church, Unified rational protein engineering with sequence-only deep representation learning. bioRxiv, 589333 (2019).

22. D. B. Goodman, G. M. Church, S. Kosuri, Causes and effects of N-terminal codon bias in bacterial genes. Science 342, 475-479 (2013).

23. J. N. Zadeh, B. R. Wolfe, N. A. Pierce, Nucleic acid sequence design via efficient ensemble defect optimization. Journal of computational chemistry 32, 439-452 (2011).

24. R. M. Dirks, M. Lin, E. Winfree, N. A. Pierce, Paradigms for computational nucleic acid design. Nucleic acids research 32, 1392-1403 (2004).

25. R. Lorenz et al., ViennaRNA Package 2.0. Algorithms for molecular biology 6, 26 (2011).

26. H. M. Salis, E. A. Mirsky, C. A. Voigt, Automated design of synthetic ribosome binding sites to control protein expression. Nature biotechnology 27, 946 (2009).

27. A. Espah Borujeni et al., Precise quantification of translation inhibition by mRNA structures that overlap with the ribosomal footprint in N-terminal coding sequences. Nucleic acids research 45, 5437-5448 (2017).

28. A. Espah Borujeni, A. S. Channarasappa, H. M. Salis, Translation rate is controlled by coupled trade-offs between site accessibility, selective RNA unfolding and sliding at upstream standby sites. Nucleic acids research 42, 2646-2659 (2013).

29. A. Espah Borujeni, H. M. Salis, Translation initiation is controlled by RNA folding kinetics via a ribosome drafting mechanism. Journal of the American Chemical Society 138, 7016-7023 (2016).

30. B. Reeve, T. Hargest, C. Gilbert, T. Ellis, Predicting translation initiation rates for designing synthetic biology. Frontiers in bioengineering and biotechnology 2, 1 (2014).

31. M. M. Meyer, The role of mRNA structure in bacterial translational regulation. Wiley Interdisciplinary Reviews: RNA 8, e1370 (2017).

32. S. Badelt, S. Hammer, C. Flamm, I. L. Hofacker, in Methods in enzymology. (Elsevier, 2015), vol. 553, pp. 193-213.

33. B. Sauerwine, M. Widom, Kinetic Monte Carlo method applied to nucleic acid hairpin folding. Physical Review E 84, 061912 (2011).

34. V. I. Jurtz et al., An introduction to deep learning on biological sequence data: examples and solutions. Bioinformatics 33, 3685-3690 (2017).

35. X.-Q. Liu, B.-X. Li, G.-R. Zeng, Q.-Y. Liu, D.-M. Ai, Prediction of Long Non-Coding RNAs Based on Deep Learning. Genes 10, 273 (2019).

36. J. Baek, B. Lee, S. Kwon, S. Yoon, Lncrnanet: long non-coding ma identification using deep learning. Bioinformatics 34, 3889-3897 (2018).

37. G. Aoki, Y. Sakakibara, Convolutional neural networks for classification of alignments of non-coding RNA sequences. Bioinformatics 34, i237-i244 (2018).

38. A. Fiannaca, M. La Rosa, L. La Paglia, R. Rizzo, A. Urso, nRC: non-coding RNA Classifier based on structural features. BioData mining 10, 27 (2017).

39. N. Frosst, G. Hinton, Distilling a neural network into a soft decision tree. arXiv preprint arXiv:1711.09784, (2017).

40. P. K. Koo, S. R. Eddy, Representation Learning of Genomic Sequence Motifs with Convolutional Neural Networks. BioRxiv, 362756 (2018).

41. K. Simonyan, A. Vedaldi, A. Zisserman, Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, (2013).

42. A. Dhawan, J. G. Scott, A. L. Harris, F. M. Buffa, Pan-cancer characterisation of microRNA across cancer hallmarks reveals microRNA-mediated downregulation of tumour suppressors. Nature communications 9, 5228 (2018).

43. Y. Xin-wei et al., STAT3 overexpression promotes metastasis in intrahepatic cholangiocarcinoma and correlates negatively with surgical outcome.