NRIP: A Model for NNRTI-RT Interaction Prediction and Enabling Virtual Screening of Anti-HIV Natural Compounds

Reverse transcriptase (RT), as an essential key enzyme in the replication process of the human immunodeficiency virus (HIV), serves as a crucial target for the development of anti-HIV drugs. Nevertheless, the frequent mutation of RT leads to drug resistance. Therefore, there is an urgent need for the rapid identification of the drug resistant-susceptible relationship. In this study, we developed the non-nucleoside RT inhibitor (NNRTI) and RT resistant-susceptible interaction prediction model (NRIP). By introducing a descriptor incorporating sequence description of RT mutation and nonuniform spatial shell structures combined with residue properties, NRIP was trained based on 4324 pairs of NNRTIs and RT interactions through an extreme gradient boosting (XGBoost) classifier. Results of 10-fold cross-validation indicated that the baseline of the sequence-descriptor-based model could reach the ROC-AUC of 0.886, which could further be increased to 0.967 by incorporating spatial descriptors. More importantly, NRIP could achieve the ROC-AUC of 0.971 and the PR-AUC of 0.974 on the independent testing dataset. Finally, a multistep virtual screening pipeline was constructed by incorporating the NRIP model with structure similarity calculation, drug likeness assessment, and molecular docking, illustrating the potential of screening bioactive compounds of natural compounds from FOODB.