Back

On the rise of AI technologies for virtual screening

Cecchini, M.; Sinenka, H.

2026-01-14 bioinformatics
10.64898/2026.01.14.699425 bioRxiv
Show abstract

AI foundational models for predicting protein-ligand interactions and binding affinities have started to emerge. We challenged Boltz-2 on a difficult dataset constructed on ten ultra-large virtual screening hit lists of pharmacologically relevant targets with in vitro binding assays. We show that Boltz-2 is the best classifier, with a success rate twice that of any other rescoring strategy. Ligand classifications by Boltz-2 are straightforward, accurate, efficient and robust, opening to million-compound accurate rankings on commodity resources.

Published in Journal of Chemical Information and Modeling (predicted rank #1) · training set

Matching journals

The top 1 journal accounts for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.