Prediction of human immunodeficiency virus type 1 drug resistance: Representation of target sequence mutational patterns via an n-grams approach

Antiretroviral medications for treating human immunodeficiency virus type 1 (HTV-1) infection, in particular inhibitors of the HTV-1 protease (PR) and reverse transcriptase (RT) enzymes, are vulnerable to the emergence of target mutations leading to drug resistance. Here we explore the relationship between PR and RT mutational patterns and corresponding changes in susceptibility to each of their eight and 11 inhibitors, respectively, by developing drug-specific predictive models of resistance trained using previously assayed and publicly available in vitro mutant data. For each inhibitor, we present tenfold cross-validation performance measures of both classification as well as regression statistical learning algorithms. Two approaches are analyzed in each case, based on the use of either relative frequencies or counts of n-grams to represent mutant protein sequences as feature vectors. To the best of our knowledge, this is the first reported study on predictive models of HTV-1 PR and RT drug resistance developed by implementing n-grams to generate sequence attributes. Our technique is complementary to other sequence-based approaches and is competitive in performance. In a novel application, we classify every pair of RT inhibitors as either potentially effective as part of a larger drug cocktail or a combination that should not be concomitantly administered, with results that closely mirror available clinical and experimental data.