The analysis of RNA structure is an important problem because RNA plays essential roles throughout all living things. Increasingly, RNA plays pivotal roles in biotechnology including RNA silencing, guide RNAs for CRISPR, ligand-binding aptamers, and drug design. While several studies have used deep learning models for analyzing RNA structure, these approaches rely on the outputs of traditional dynamic programming algorithms based on known thermodynamic energy parameters during training. The challenge remains to train neural networks to independently learn thermodynamic principles, so that potentially new biology can be uncovered. Here we present DeepSloop, a deep recurrent network capable of distinguishing hairpin-forming sequences from non-hairpin-forming, and that is trained purely on class-labeled nucleotide sequences. We demonstrate that DeepSloop can accurately classify hairpin stem-loops, and thus learned a distinct internal representation of structure that includes destabilizing energies of bulges and internal loops.

MOTIVATION: Predicting the secondary structure of an ribonucleic acid (RNA) sequence is useful in many applications. Existing algorithms [based on dynamic programming] suffer from a major limitation: their runtimes scale cubically with the RNA length, and this slowness limits their use in genome-wide applications. RESULTS: We present a novel alternative O(n3)-time dynamic programming algorithm for RNA folding that is amenable to heuristics that make it run in O(n) time and O(n) space, while producing a high-quality approximation to the optimal solution. Inspired by incremental parsing for context-free grammars in computational linguistics, our alternative dynamic programming algorithm scans the sequence in a left-to-right (5'-to-3') direction rather than in a bottom-up fashion, which allows us to employ the effective beam pruning heuristic. Our work, though inexact, is the first RNA folding algorithm to achieve linear runtime (and linear space) without imposing constraints on the output structure. Surprisingly, our approximate search results in even higher overall accuracy on a diverse database of sequences with known structures. More interestingly, it leads to significantly more accurate predictions on the longest sequence families in that database (16S and 23S Ribosomal RNAs), as well as improved accuracies for long-range base pairs (500+ nucleotides apart), both of which are well known to be challenging for the current models. AVAILABILITY AND IMPLEMENTATION: Our source code is available at https://github.com/LinearFold/LinearFold, and our webserver is at http://linearfold.org (sequence limit: 100 000nt). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.