A deep learning model for predicting next-generation sequencing depth from DNA sequence.

Targeted high-throughput DNA sequencing is a primary approach for genomics and molecular diagnostics, and more recently as a readout for DNA information storage. Oligonucleotide probes used to enrich gene loci of interest have different hybridization kinetics, resulting in non-uniform coverage that increases sequencing costs and decreases sequencing sensitivities. Here, we present a deep learning model (DLM) for predicting Next-Generation Sequencing (NGS) depth from DNA probe sequences. Our DLM includes a bidirectional recurrent neural network that takes as input both DNA nucleotide identities as well as the calculated probability of the nucleotide being unpaired. We apply our DLM to three different NGS panels: a 39,145-plex panel for human single nucleotide polymorphisms (SNP), a 2000-plex panel for human long non-coding RNA (lncRNA), and a 7373-plex panel targeting non-human sequences for DNA information storage. In cross-validation, our DLM predicts sequencing depth to within a factor of 3 with 93% accuracy for the SNP panel, and 99% accuracy for the non-human panel. In independent testing, the DLM predicts the lncRNA panel with 89% accuracy when trained on the SNP panel. The same model is also effective at predicting the measured single-plex kinetic rate constants of DNA hybridization and strand displacement.

High-throughput methods for measuring DNA thermodynamics.

Understanding the thermodynamics of DNA motifs is important for prediction and design of probes and primers, but melt curve analyses are low-throughput and produce inaccurate results for motifs such as bulges and mismatches. Here, we developed a new, accurate and high-throughput method for measuring DNA motif thermodynamics called TEEM (Toehold Exchange Energy Measurement). It is a refined framework of comparing two toehold exchange reactions, which are competitive strand displacement between oligonucleotides. In a single experiment, TEEM can measure over 1000 ΔG° values with standard error of roughly 0.05 kcal/mol.

Nucleic Acid Quantitation with Log-Linear Response Hybridization Probe Sets.

Concentrations of different nucleic acid species in biological samples span many orders of magnitude. A real-time polymerase chain reaction maps the concentration of a target nucleic acid sequence log-linearly into cycle threshold to enable quantitation with a wide dynamic range but suffers from enzymatic biases. Here, we present a general design for constructing hybridization probe sets with highly log-linear response curves to enable accurate enzyme-free quantitation across large ranges (more than 6 logs) of target DNA concentrations. The sensitivity of each component probe is accurately adjusted via formulation stoichiometry to reduce the standard error of target quantitation down to 7%. As a proof of concept, we show multiplexed quantitation of three microRNA species in total RNA of the human brain and liver.

Predicting DNA hybridization kinetics from sequence.

Hybridization is a key molecular process in biology and biotechnology, but so far there is no predictive model for accurately determining hybridization rate constants based on sequence information. Here, we report a weighted neighbour voting (WNV) prediction algorithm, in which the hybridization rate constant of an unknown sequence is predicted based on similarity reactions with known rate constants. To construct this algorithm we first performed 210 fluorescence kinetics experiments to observe the hybridization kinetics of 100 different DNA target and probe pairs (36 nt sub-sequences of the CYCS and VEGF genes) at temperatures ranging from 28 to 55 °C. Automated feature selection and weighting optimization resulted in a final six-feature WNV model, which can predict hybridization rate constants of new sequences to within a factor of 3 with ∼91% accuracy, based on leave-one-out cross-validation. Accurate prediction of hybridization kinetics allows the design of efficient probe sequences for genomics research.

Continuously tunable nucleic acid hybridization probes.

In silico-designed nucleic acid probes and primers often do not achieve favorable specificity and sensitivity tradeoffs on the first try, and iterative empirical sequence-based optimization is needed, particularly in multiplexed assays. We present a novel, on-the-fly method of tuning probe affinity and selectivity by adjusting the stoichiometry of auxiliary species, which allows for independent and decoupled adjustment of the hybridization yield for different probes in multiplexed assays. Using this method, we achieved near-continuous tuning of probe effective free energy. To demonstrate our approach, we enforced uniform capture efficiency of 31 DNA molecules (GC content, 0-100%), maximized the signal difference for 11 pairs of single-nucleotide variants and performed tunable hybrid capture of mRNA from total RNA. Using the Nanostring nCounter platform, we applied stoichiometric tuning to simultaneously adjust yields for a 24-plex assay, and we show multiplexed quantitation of RNA sequences and variants from formalin-fixed, paraffin-embedded samples.