Recent computational advances in the identification of cryptic binding sites for drug discovery

Cryptic ligand binding sites, defined as binding pockets that exist in the ligand-bound state of a protein but not in its apo form, are gaining increasing interest due to the opportunities they provide for drug discovery.

This review article looks at the current state of cryptic binding site research, highlighting advancements in both molecular dynamic (MD) methods and machine learning (ML) methods to predict and utilize these sites.

MD methods include the use of Markov State Models, Enhanced Sampling, and other methods such as Cosolvent MD, while ML methods utilize Support Vector Machine, Random Forest, and Neural Networks. Here, we discuss case studies for both methods and their overlaps, providing insight into the future and the limitations faced. Compared to MD methods, ML methods are often reported to be more cost- and time-effective. However, a limited number of datasets are available for training these ML methods. Integrating MD with ML methods promises to expand our ability to predict and validate new cryptic binding sites that can be evaluated for druggability.