COMP 57

An automated non-linear learning model extracts the key binding patterns from the chemical interaction properties encoded in the ISPT descriptor value vector of a set of compounds with known biological activity. The acquired knowledge is applied to evaluate the surface interaction models of the screened chemical compounds. The advantage of the ISPT descriptor of eHiTS LASSO over the 2D fingerprint based descriptors is the independence from the underlying 2D structural motifs allowing the recognition of structurally diverse ligands with similar interaction profiles. On the other hand, the eHiTS LASSO ISPT descriptor does not dependent on the 3D shape of the surface, thus the descriptor values are independent of the conformation of the ligand, which is an advantage over other surface based descriptors that are biased by the specific input conformation.

The method has been validated on a variety of practical virtual screening tasks. Results will be presented demonstrating the ability of the software to retrieve the majority of actives from the screening essay at the top of the ranking list. Cluster analysis using traditional 2D structural descriptors is used to highlight the scaffold differences in the actives retrieved by eHiTS LASSO with high ISPT similarity scores. The scaffold hopping ability of the descriptor makes LASSO a powerful tool for finding new leads without the same toxicity or potential intellectual property issues as the query.