CNN for human exons and introns classification

Abstract

Modeling properties and functions associated with DNA sequences is a very complex task in genomics. This task is more and more difficult when it concerns the human genome. This genome is composed with coding and non-coding regions which are not yet fully identified. Difficulties seen in this context are particularly related to the fact that 98% of the human genome is made up of non-coding zones. Therefore it seems evident that a powerful predictive model can have a huge advantage in advancing the exploration of the human genome. In this paper, we started from text representation of the DNA sequence to the representation of exons and introns by images to classify them. Here we are introducing a convolutional neural network model to classify human exons and introns. Our model has shown very good results concerning classification learning and testing rate which exceed 90%.