Site Identification by Ligand Competitive Saturation (SILCS) is a multiple solute-solvent simulation method for determining functional group affinity patterns for proteins. Using a grand canonical Monte Carlo with molecular dynamics (GCMC-MD) simulation approach, SILCS drives efficient sampling of chemically-diverse probes and water molecules across a protein surface and into deeply buried pockets. The resulting trajectories are processed to create three-dimensional affinity maps (FragMaps) for the different probe atoms. SILCS FragMaps recapitulate functional group patterns of crystallographic ligands across diverse targets, such as nuclear receptors and GPCRs from different gene families, without any target-specific training. FragMap-based Ligand Grid Free Energy (LGFE) is a scoring metric that rapidly rank-orders ligand favorability for any region of a protein. In our recent validation effort with 8 protein targets and 407 ligands with broad chemical variability, it was shown that LGFE scoring can predict the experimental free energy change in 77% of the times. Recent development of machine learning algorithms and deep learning algorithms can further enhance the utility of SILCS based technology in CADD. For example, the LGFE scoring can be further improved by machine learning algorithms, which can improve the accuracy to 82% in the same benchmark. In addition, a deep learning algorithm is being developed to predict the FragMap, which could greatly improves the availability of FragMap without a lengthy simulation. Finally, active learning based virtual screening will be discussed, which could tackle screening huge library of chemical database greater than 10<sup>9</sup> compounds.

Computer-aided drug discovery using SILCS technology and AI-assisted screening of ultra-large compound library

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.