Improving deep models of protein-coding potential with a Fourier-transform architecture and machine translation task.

Ribosomes are information-processing macromolecular machines that integrate complex sequence patterns in messenger RNA (mRNA) transcripts to synthesize proteins. Studies of the sequence features that distinguish mRNAs from long noncoding RNAs (lncRNAs) may yield insight into the information that directs and regulates translation. Computational methods for calculating protein-coding potential are important for distinguishing mRNAs from lncRNAs during genome annotation, but most machine learning methods for this task rely on previously known rules to define features. Sequence-to-sequence (seq2seq) models, particularly ones using transformer networks, have proven capable of learning complex grammatical relationships between words to perform natural language translation. Seeking to leverage these advancements in the biological domain, we present a seq2seq formulation for predicting protein-coding potential with deep neural networks and demonstrate that simultaneously learning translation from RNA to protein improves classification performance relative to a classification-only training objective. Inspired by classical signal processing methods for gene discovery and Fourier-based image-processing neural networks, we introduce LocalFilterNet (LFNet). LFNet is a network architecture with an inductive bias for modeling the three-nucleotide periodicity apparent in coding sequences. We incorporate LFNet within an encoder-decoder framework to test whether the translation task improves the classification of transcripts and the interpretation of their sequence features. We use the resulting model to compute nucleotide-resolution importance scores, revealing sequence patterns that could assist the cellular machinery in distinguishing mRNAs and lncRNAs. Finally, we develop a novel approach for estimating mutation effects from Integrated Gradients, a backpropagation-based feature attribution, and characterize the difficulty of efficient approximations in this setting.

Improving deep models of protein-coding potential with a Fourier-transform architecture and machine translation task.

Ribosomes are information-processing macromolecular machines that integrate complex sequence patterns in messenger RNA (mRNA) transcripts to synthesize proteins. Studies of the sequence features that distinguish mRNAs from long noncoding RNAs (lncRNAs) may yield insight into the information that directs and regulates translation. Computational methods for calculating protein-coding potential are important for distinguishing mRNAs from lncRNAs during genome annotation, but most machine learning methods for this task rely on previously known rules to define features. Sequence-to-sequence (seq2seq) models, particularly ones using transformer networks, have proven capable of learning complex grammatical relationships between words to perform natural language translation. Seeking to leverage these advancements in the biological domain, we present a seq2seq formulation for predicting protein-coding potential with deep neural networks and demonstrate that simultaneously learning translation from RNA to protein improves classification performance relative to a classification-only training objective. Inspired by classical signal processing methods for gene discovery and Fourier-based image-processing neural networks, we introduce LocalFilterNet (LFNet). LFNet is a network architecture with an inductive bias for modeling the three-nucleotide periodicity apparent in coding sequences. We incorporate LFNet within an encoder-decoder framework to test whether the translation task improves the classification of transcripts and the interpretation of their sequence features. We use the resulting model to compute nucleotide-resolution importance scores, revealing sequence patterns that could assist the cellular machinery in distinguishing mRNAs and lncRNAs. Finally, we develop a novel approach for estimating mutation effects from Integrated Gradients, a backpropagation-based feature attribution, and characterize the difficulty of efficient approximations in this setting.

LtrDetector: A tool-suite for detecting long terminal repeat retrotransposons de-novo.

BACKGROUND: Long terminal repeat retrotransposons are the most abundant transposons in plants. They play important roles in alternative splicing, recombination, gene regulation, and defense mechanisms. Large-scale sequencing projects for plant genomes are currently underway. Software tools are important for annotating long terminal repeat retrotransposons in these newly available genomes. However, the available tools are not very sensitive to known elements and perform inconsistently on different genomes. Some are hard to install or obsolete. They may struggle to process large plant genomes. None can be executed in parallel out of the box and very few have features to support visual review of new elements. To overcome these limitations, we developed LtrDetector, which uses techniques inspired by signal-processing. RESULTS: We compared LtrDetector to LTR_Finder and LTRharvest, the two most successful predecessor tools, on six plant genomes. For each organism, we constructed a ground truth data set based on queries from a consensus sequence database. According to this evaluation, LtrDetector was the most sensitive tool, achieving 16-23% improvement in sensitivity over LTRharvest and 21% improvement over LTR_Finder. All three tools had low false positive rates, with LtrDetector achieving 98.2% precision, in between its two competitors. Overall, LtrDetector provides the best compromise between high sensitivity and low false positive rate while requiring moderate time and utilizing memory available on personal computers. CONCLUSIONS: LtrDetector uses a novel methodology revolving around k-mer distributions, which allows it to produce high-quality results using relatively lightweight procedures. It is easy to install and use. It is not species specific, performing well using its default parameters on genomes of varying size and repeat content. It is automatically configured for parallel execution and runs efficiently on an ordinary personal computer. It includes a k-mer scores visualization tool to facilitate manual review of the identified elements. These features make LtrDetector an attractive tool for future annotation projects involving long terminal repeat retrotransposons.