Optimized Virtual Screening Workflow: Towards Target-Based Polynomial Scoring Functions for HIV-1 Protease

Background: One key step in the development of inhibitors for an enzyme is the application of computational methodologies to predict protein-ligand interactions. The abundance of structural and ligand-binding information for HIV-1 protease opens up the possibility to apply computational methods to develop scoring functions targeted to this enzyme. Objective: Our goal here is to develop an integrated molecular docking approach to investigate protein-ligand interactions with a focus on the HIV-1 protease. In addition, with this methodology, we intend to build target-based scoring functions to predict inhibition constant (Ki) for ligands against the HIV-1 protease system. Methods: Here, we described a computational methodology to build datasets with decoys and actives directly taken from crystallographic structures to be applied in evaluation of docking performance using the program SAnDReS. Furthermore, we built a novel function using as terms MolDock and PLANTS scoring functions to predict binding affinity. To build a scoring function targeted to the HIV-1 protease, we have used machine-learning techniques. Results: The integrated approach reported here has been tested against a dataset comprised of 71 crystallographic structures of HIV protease, to our knowledge the largest HIV-1 protease dataset tested so far. Comparison of our docking simulations with benchmarks indicated that the present approach is able to generate results with improved accuracy. Conclusion: We developed a scoring function with performance higher than previously published benchmarks for HIV-1 protease. Taken together, we believe that the approach here described has the potential to improve docking accuracy in drug design projects focused on HIV-1 protease.