Multilayer perceptron and evolutionary radial basis function neural network models for discrimination of HIV-1 genomes

Abstract

High rate of mutation and frequent recombination cause evolution of HIV-1 very diverse and adaptive. Revealing the recombination patterns in HIV-1 is a computationally intensive problem. Techniques based on phylogenetic analysis are not suitable for genomelevel studies. Here we elucidate approaches based on multilayer perceptron and evolutionary radial basis function neural network for the analysis of 4130 HIV- 1 genomes. These techniques show remarkable improvement over other machine learning techniques used for such classification. The models outperformed other machine learning models having 92% classification accuracy. Multilayer perceptron achieved sensitivity and specificity of 82% and 96%, whereas radial basis function neural network achieved sensitivity and specificity of 78% and 98% on tenfold cross-validation respectively.