Can predicted protein structures help us find new inhibitors?

Protein–protein interfaces are attractive but challenging drug targets. Jordi Gómez Borrego and Marc Torrent Burgas compared eight docking protocols across 16 interactions with known modulators, using both experimental structures and AlphaFold2 models.

The predicted models performed similarly to the experimental structures in this benchmark. Focusing docking on a known binding region worked better than searching the entire protein, with TankBind_local and Glide among the strongest performers. Molecular dynamics refinement did not deliver a consistent improvement across protocols. These findings help guide how we use structural predictions to prioritise compounds for testing.

Published in Journal of Cheminformatics.