Residual Neural Network in Genomics

Abstract

Residual neural network (ResNet) is a Deep Learning model introduced by He et al. in 2015 to enhance traditional convolutional neural networks proposed to solve computer vision problems. It uses skip connections over some layer blocks to avoid vanishing gradient problem. Currently, many researches are focused to test and prove the efficiency of the ResNet on different domains such as genomics. In fact, the study of human genomes provides important information on the detection of diseases and their best treatments. Therefore, most of the scientists opted for bioinformatics solutions to get results in a reasonable time. In this paper, our interest is to show the effectiveness of the ResNet model on genomics. For that, we propose two new ResNet models to enhance the results of two genomic problems previously resolved by CNN models. The obtained results are very promising and they proved the performance of our ResNet models compared to the CNN models.