End-to-end deep model for pre-miRNA prediction

RAAD, J; STEGMAYER, G; BUGNON, L. A.; MILONE, D H

virtual

Congreso; XICAB2C; 2021

A2B2C

MicroRNAs (miRNAs) are small RNA sequences with key roles in the regulation of gene expression at post-transcriptional level in different species. Accurate prediction of novel miRNAs is needed due to their importance in many biological processes and their associations with complicated diseases. Recently, several deep learning models for prediction of new miRNAs have been proposed, outperforming models that are based on handcrafted features. However, even these deep learning approaches require complex pre-processing of the input sequences and the prediction of secondary structures in order to reach an acceptable performance. In order to eliminate these pre-processing it is necessary to design a structure prediction model that can transfer the learned characteristics. In this way, it is possible to obtain an end-to-end model capable of predicting new miRNAs from the input sequence only.In this work we present the first full end-to-end deep learning model for pre-miRNA prediction. This model is based on Transformers, a novel neural architecture that uses attention mechanisms to infer global dependencies between inputs and outputs. It is capable of receiving the raw genome-wide data as input, without any pre-processing or feature extraction. The model has been validated through several experimental setups using the human genome, and it was compared with state-of-the-art algorithms obtaining 10 times better performance.