A Machine Learning approach for pre-miRNA discovery in SARS-CoV-2

MERINO, GABRIELA A; BUGNON, L. A.; RAAD, J; ARIEL, FEDERICO; MILONE, D H; STEGMAYER, G

Conferencia; ISMB/ECCB 2021; 2021

ISCB

We have developed a novel approach based on machine learning (ML) for identifying precursors of microRNAs (pre-miRNAs) in the genome of the novel coronavirus SARS-CoV-2. The discovery of miRNAs in the novel virus is of high importance in the context of the current sanitary crisis for the improvement of diagnostic and treatment strategies. For the discovery of pre-miRNAs 3 ML methods were used in combination: a novel deep convolutional neural network (mirDNN), a deep self-organizing map (deeSOM), and a one-class support vector machine (OC-SVM). Each method provided a list of candidates to potential pre-miRNAs in the viral genome, supported by a score. In this study, pre-miRNAs were identified as those having scores in the top 10th percentile in all methods. With this approach, 12 candidate structures were discovered in the viral genome and validated with small RNA-seq data. The expression of 8 mature miRNAs-like sequences was confirmed from SARS-CoV-2 infected human cells. The predicted miRNAs were found as targeting a subset of human genes of which 109 are transcriptionally deregulated upon infection, and 28 of those genes are down-regulated in infected human cells and related to respiratory diseases and viral infection, previously associated with SARS-CoV-1 and SARS-CoV-2.