miRe2e: a full end-to-end deep model based on Transformers for prediction of pre-miRNAs

RAAD, JONATHAN; BUGNON, LEANDRO A; MILONE, DIEGO H; STEGMAYER, GEORGINA

BIOINFORMATICS (OXFORD, ENGLAND)

OXFORD UNIV PRESS

Año: 2022 vol. 38

1367-4803

Motivation: MicroRNAs (miRNAs) are small RNA sequences with key roles in the regulation of geneexpression at post-transcriptional level in different species. Accurate prediction of novel miRNAs is neededdue to their importance in many biological processes and their associations with complicated diseasesin humans. Many machine learning approaches were proposed in the last decade for this purpose, butrequiring handcrafted features extraction in order to identify possible de novo miRNAs. More recently, theemergence of deep learning has allowed the automatic feature extraction, learning relevant representationsby themselves. However, the state-of-art deep models require complex pre-processing of the inputsequences and prediction of their secondary structure in order to reach an acceptable performance.Results: In this work we present miRe2e, the first full end-to-end deep learning model for pre-miRNAprediction. This model is based on Transformers, a neural architecture that uses attention mechanismsto infer global dependencies between inputs and outputs. It is capable of receiving the raw genome-widedata as input, without any pre-processing nor feature engineering. After a training stage with known premiRNAs, hairpin and non-harpin sequences, it can identify all the pre-miRNA sequences within a genome.The model has been validated through several experimental setups using the human genome, and it wascompared with state-of-the-art algorithms obtaining 10 times better performance.Availability and Implementation: Webdemo available at https://sinc.unl.edu.ar/web-demo/miRe2e/ andsource code available for download at https://github.com/sinc-lab/miRe2e