DNA Genome Classification with Machine Learning and Image Descriptors

Sequence alignment is the most used method in Bioinformatics. Nevertheless, it is slow in time processing. For that reason, there are several methods not based on alignment to compare sequences. In this work, we analyzed Kameris and Castor, two alignment-free methods for DNA genome classification; we compared them against the most popular CNN networks: VGG16, VGG19, Resnet-50, and Inception. Also, we compared them with image descriptor methods like First-order Statistics(FOS), Gray-level Co-occurrence matrix (GLCM), Local Binary Pattern (LBP), and Multi-resolution Local Binary Pattern(MLBP), and classifiers like: Support Vector Machine (SVM), Random Forest (RF) and k-nearest neighbors (KNN). In this comparison, we concluded that FOS, GLCM, LBP, and MLBP, all with SVM got the best results in f1-score, followed by Castor and Kameris and finally by CNNs. Furthermore, Castor got a minor processing time. Finally, according to experiments, 5-mer (used by Kameris and Castor) and 6-mer outperformed 7-mer.