Deep Learning Based Phylogenetic Analysis

Abstract

The classification made by looking at homologous organs to understand the evolutionary relationship between various taxonomic groups is called phylogenetic classification. When classifying living species, features such as genetic information, origin similarity, degree of kinship, homologous organs of organisms and DNA sequences are considered. DNA sequences contain the most important information among these features. The basic method for phylogenetics is the extraction of DNA sequences, the formation of the phylogenetic tree and the understanding of the class to which the species belongs. In this study, a method for phylogenetic analysis is proposed by classifying DNA sequences of five different types using two different numerical mapping techniques. DNA sequences of 5 different types were converted into digital signals using Entropy-based numerical mapping and EIIP techniques. The feature extraction was made by using ResNet, which is one of the ESA models, from the digitized DNA sequences. With the feature extraction using ESA models, features are automatically obtained without the need for manual feature extraction from raw data. The obtained properties were then classified by the Support Vector Machine (SVM) and the k-Nearest neighbor algorithm (k-NN). As a result, DNA gene sequences belonging to five different species were classified with 93.60% accuracy.