Assessing ENM-NMA based molecular descriptors of HIV-1 protease for drug resistance prediction by machine learning methods

Resumen

Drug resistance is a major factor in the failure of antiretroviral therapy against human immunodeficiency virus (HIV). Due to the high costs of direct phenotypic assays to assess drug resistance, genotypic assays, based on sequencing of the viral genome or part of it, are commonly used to infer drug resistance. For genotypic testing, interpretation of sequence information is the biggest challenge. The large amount of data linking genotype and phenotype information provided a framework for predicting drug resistance from genotype, based on machine learning methods. Current methods rely primarily on sequence information from observed variants and still largely fail to predict resistance in previously unobserved variants. The inclusion of structural and dynamic information is supposed to improve predictions. The use of molecular descriptors based on dynamic information has been limited by their computational cost of calculation. This study shows the feasibility of dynamic descriptors derived from normal mode analysis in elastic network models of HIV-1 protease for predicting drug resistance. Publicly available HIV-1 sequence data and drug resistance assay results for 7 ART drugs were used to evaluate the performance of 4 prediction algorithms using classical and dynamic descriptors separately, obtaining comparable results.