Structure-based Virtual Screening (VS)1,2 aims at identifying compounds with previously unknown affinity for a target from its three-dimensional (3D) structure. Docking techniques are typically used to carry out this in silico prediction using their embedded scoring functions (SFs). When applied to VS, SFs seek to rank compounds based on their predicted affinity for the target as a way to discriminate between binders and non-binders. Despite the well-known limitations of SFs1,3,4,5, their application has been beneficial in many VS projects and successful applications have been reported3,6,7,8,9. Although the classical SFs used in VS experiments have often proven useful, improved accuracy requires novel approaches.