A Machine Learning Approach to Enhance Scoring Performance in Docking-Based Virtual Screening Experiments: COX-1 as a Case Study

Molecular docking can be reasonably successful at reproducing X-ray poses of a ligand in the binding site of a protein. However, scoring functions are typically unsuccessful at correctly ranking ligands according to their binding affinity. Using cyclooxygenase-1 (COX-1), a particularly challenging workhorse in virtual screening (VS) we show how the use of support vector machines (SVMs), trained with the individual energy terms retrieved from docking-based VS experiments, can improve the discrimination between active and inactive compounds. Actives and inactives for COX-1 were obtained from the Directory of Useful Decoys (DUD) and docked into COX-1 with AutoDock Vina (Vina). The energy parameters of Vina’s scoring function were used to train classification models with SVM-light. The results show that Vina offers acceptable pose prediction accuracy, but its scoring function performs poorly compared to our SVM classification models. The superior performance of the trained classification models highlights the potential of using non-linear machine learning methods to identify bioactive compounds through docking-based screening.