Virtual screening models for prediction of HIV-1 RT associated RNase H inhibition

The increasing resistance to current therapeutic agents for HIV drug regiment remains a major problem for effective acquired immune deficiency syndrome (AIDS) therapy. Many potential inhibitors have today been developed which inhibits key cellular pathways in the HIV cycle. Inhibition of HIV-1 reverse transcriptase associated ribonuclease H (RNase H) function provides a novel target for anti-HIV chemotherapy. Here we report on the applicability of conceptually different in silico approaches as virtual screening (VS) tools in order to efficiently identify RNase H inhibitors from large chemical databases. The methods used here include machine-learning algorithms (e.g. support vector machine, random forest and kappa nearest neighbor), shape similarity (rapid overlay of chemical structures), pharmacophore, molecular interaction fields-based fingerprints for ligands and protein (FLAP) and flexible ligand docking methods. The results show that receptor-based flexible docking experiments provides good enrichment (80-90%) compared to ligand-based approaches such as FLAP (74%), shape similarity (75%) and random forest (72%). Thus, this study suggests that flexible docking experiments is the model of choice in terms of best retrieval of active from inactive compounds and efficiency and efficacy schemes. Moreover, shape similarity, machine learning and FLAP models could also be used for further validation or filtration in virtual screening processes. The best models could potentially be use for identifying structurally diverse and selective RNase H inhibitors from large chemical databases. In addition, pharmacophore models suggest that the inter-distance between hydrogen bond acceptors play a key role in inhibition of the RNase H domain through metal chelation.