Plants’ miRNAs identification from deep-sequencing RNA-seq data using a multi-layer perceptron

Micro-RNA (miRNA) – mediated transcript degradation is a layer of gene regulation at the post-transcriptional level that has important roles in plants. Some traits of plants that are of interest to the food industry are tightly regulated by this molecular mechanism. In México, Fabaceae – family plants represent one of the main food sources. Accordingly, it is important to study this layer of regulation to improve crop and seed production yields, nonetheless, one of the pressing concerns is the miRNAs loci identification. The basic and ancillary criteria, sometimes are not enough evidence for identifying miRNA loci. Artificial intelligence (AI), such as convolutional neural networks (CNN), have shown excellent predic-tive performance in identifying miRNAs loci, however, some of these CNN are complex and difficult to train and run. A multi-layer perceptron (MLP) model has been proposed for identifying pre-miRNAs sequences; it processes 180 feature information, however, the analysis is limited by the feature calculation, because it is computationally intensive. In this work, we proposed the use of AI based on a multi-layer perceptron (MLP) model which isn’t complex and easy to train, we also pro-posed the use of k-mer frequencies to extract information from nucleotide and secondary-structure representation sequences. We tested several features of MLP models such as activation functions between layers and the number of dropout layers. The best-fitted mo-dels showed 84-90% of sensitivity and 98 to 100% of specificity when they were evaluated with testing datasets. We tested the predictive performance of the best-fitted models on real deep RNA-seq data. In conclusion, in this paper, we present an MLP-based AI capable of identifying pre-miRNAs sequences from the Fabaceae family plants using deep RNA-Seq data, these AIs showed sensitivity values of 80-85% and specificity values of 90-95%.