Enriched DNA Strands Classification using CGR Images and Convolutional Neural Network

Abstract

Bioinformatics is the biological study which applies programming techniques for more understanding and analysis of living objects such as the study of genome structure. The genome structure could be represented in the form of an image. Chaos Game Representation (CGR) is the practice of converting the DNA sequence (i.e., genomes) into images, where each image is a graphical appearance for an individual DNA strand's signature. CGR is a method of converting a long one-dimensional DNA sequence into a graphical form. This method provides a visual image of a DNA sequence different from the traditional manual linear arrangement of nucleotides polymerase chain reaction. In the recent years, CGR was introduced to automatically classify genomes not only by archival references but also through its' unique signature. In this paper, a novel CGR classification approach is developed combining the advances of image processing and pattern recognition approaches. The approach starts by declaring the genome and using the CGR technique to map it to the graphical interface (i.e., 16x16 signature images). Then, an image processing procedure is prepared to handle complex geometric shapes, analyze structured and visualized genome sequences and fractal point the included nucleotides of these images. Finally, the convolutional neural network was designed and well-trained by those signatures to classify each genome tested.