Difference in Virtual Screening Enrichment Using polar and non-polar interactions of MM-PB/GBSA methods

Over-simplification has caused trouble less than trivial within virtual screening using scoring functions. Previous studies have sought solutions using the binding free energy re-evaluation for outcome refinement. [1-6] In this work, we used an efficient docking scheme, PLANTS, [7] in combination with the MM-PBSA method implemented using AMBER package for rescoring. [8] This resulted in a workflow suitable for high throughput applications. We applied an ensemble evaluation strategy, by using the average score over multiple similar best-ranked poses from the docking procedures. This was intended to mimic the dynamic behaviour of ligands within the binding sites. We tested the rescoring scheme using three test sets consisting of proteins from different classes, each with a size around 20 protein targets. We performed extensive investigations on the effects from using different internal dielectric constants (1, 2, 4, and 6) on virtual screening performance. We noticed that the improvements were not universal, as multiple systems shown worsen or invariant performance when increasing the internal dielectric value, contradicting previous studies [5]. We thus analysed the performance of different energy compositions separately. We noticed the non-polar energy calculated from MM-PB/GBSA are more likely to provide the best virtual screening outcomes. However, the electrostatic compositions provided the devastating effects in majority of the tested systems. Moreover, increasing the internal dielectric values more likely to decrease the effects from electrostatic energy in the final binding free energy estimation, hence, resulting in the improvements observed in several systems. We suspect that the poor capability of electrostatic energies to distinguish actives from inactive compounds inevitably affects the performance of MM-PBSA method in virtual screening. Further improvements, either in parameterisation or using alternative numerical methods, might provide optimal outcomes.