HIV drug resistance prediction using multiple regression

Drug resistance is commonly encountered during treatment for HIV/AIDS, and decreases the efficacy of the antiviral drugs. Genotyping the infecting virus gives sequence data for computational prediction of resistance, which is more efficient than performing experimental assays for resistance. Current predictions rely on simple rules with modest accuracy; therefore, a prediction method with high accuracy is needed to improve drug selection for therapy. Here, we apply a hybrid sequence/structure protein representation in conjunction with multiple regression for predicting resistance to drugs. The algorithm was tested on genotype-phenotype data for HIV-1 protease (PR) and HIV-1 reverse transcriptase (RT). The overall cross-validated regression R²-values were 0.51-0.72 for predicting resistance to four PR inhibitors; and 0.76-0.91 for three RT inhibitors demonstrating successful predictions.