Some time back, John Van Drie and myself had done some work on characterizing structure-activity cliffs, which are molecules that have very similar structures but very different activities. The term originated from Maggiora, who suggested that this was a reason for the failure of many QSAR models. At the same time, such cliffs can represent the most interesting portions of a structure-activity relationship. Our first paper introduced the idea of the Structure Activity Landscape Index (SALI) – a numerical metric to characterize activity cliffs and SAR landscapes, and was followed by a paper describing how one could use SALI to characterize the quality of predictive models. We also put out some software to let others use SALI to analyse their own data.
A few weeks back I heard from Jinhua Zhang at Simulations Plus Inc., that they had implented the SALI methodology in their ADMET Predictor product. He also provided some slides showing how they used SALI to characterize a variety of predictive models for hERG inhibition.
It’s a nice feeling to see our theoretical work being incorporated into a real world product!