Prediction of integrase inhibitor resistance from genotype through neural network modeling

HIV-1 integrase (IN) has become an important target for antiviral drug discovery. While AIDS drug treatment often fails due to the emergence of drug resistant species. Elvitegravir (EVG) is one of the FDA-approved drugs. We developed a neural network prediction model to make a qualitative EVG resistance phenotype prediction. First, we developed a genotype-phenotype database. Secondly, we classified the multiple mutations at the same site in three different ways: mutations result in the same volume change, the same charge change or both the same volume and charge changes. Finally, we proposed three neural network models based on the above three different ways of classification. The results show that the prediction accuracy of volume model over the training set and test set are 92.2% and 91.8%, respectively. The drug susceptibility of new mutant strains to EVG can be predicted using this model, and the model can be applied as a diagnostic service for clinicians.