New SILCS-MC Study Demonstrates Improved Performance with Machine Learning Optimization


New SILCS-MC Study Demonstrates Improved Performance with Machine Learning Optimization 


May 16, 2019, Baltimore, MDSilcsBio, a company that develops commercial software and related services for structure-based drug design, today announced that established Site-identification by ligand competitive saturation-Monte Carlo (SILCS-MC) approaches were applied to seven protein targets and 551 ligands, correctly predicting relative affinities an average of 69% of the time.

A new target-specific machine-learning (ML) optimization of the SILCS-MC scoring function was able to further improve the predictive capabilities using only a small set of experimental binding data (about 30 compounds). The ML optimization increased predictability to 76%, with predictability rates of over 80% for several data sets.

“We’re excited that our ML approach was able to improve predictability by coupling experimental data to the physically correct SILCS FragMaps,” says Sunhwan Jo, SilcsBio’s Director of Commercial Development. “This is an interesting new direction for us.”

The extensive validation study from researchers at University of Maryland, Baltimore is published in the Journal of Chemical Information and Modeling (JCIM) and is titled “Optimization and Evaluation of Site-Identification by Ligand Competitive Saturation (SILCS) as a Tool for Target-Based Ligand Optimization.

Notably, the 76% correct relative prediction rate of SILCS for the 7 targets in this study is similar to or better than the industry standard Free Energy Perturbation (FEP) method while being significantly faster. This is because the compute intensive SILCS FragMap generation is performed only once per protein target, while FEP requires separate calculations for each ligand being evaluated.

In addition, while FEP is restricted to small ligand modifications, SILCS protocols can accommodate ligands in a congeneric series, scaffold changes, and a variety of larger chemical modifications such as ring additions. The results further support the utility of SILCS as a powerful and computationally accessible tool to support lead compound optimization in drug discovery.

About SILCS:

SILCS is a cosolvent sampling technique that provides FragMaps, 3-D maps of the binding affinity pattern of chemical fragments on a protein. These FragMaps are leveraged in computer-aided drug design (CADD) for a variety of purposes, including rapid evaluation of ligands via SILCS-MC, a Monte Carlo sampling technique to predict binding conformations and relative binding affinities. SILCS and SILCS-MC provide a much faster alternative to FEP methods, the current CADD industry standard, as FragMaps do not need to be recomputed for each ligand.

About SilcsBio, LLC:

SilcsBio started operations in April 2013 based on licensed intellectual property developed by Dr. Alex MacKerell at the University of Maryland, Baltimore where he is the Grollman-Glick Professor of Pharmaceutical Sciences and Director of the Computer-Aided Drug Design Center. SilcsBio offers commercial software and related services for structure-based drug design and has customers in the US, UK, Japan, and Europe. The company is headquartered at 8 Market Place, Suite 300, Baltimore, MD 21202.

Media Contact:

Tina Guvench
Technical Marketing Manager