NYU researchers release tool for predicting drug molecules' hydrogen positions

New York University researchers have released an open-source tool that predicts which hydrogen positions are stable in drug-like molecules. In one test, it processed about 4.6 million compounds in 3.2 hours on a single GPU-enabled node. A hydrogen atom can move between sites in a molecule, changing how that molecule may interact with a protein. The researchers trained their model on more than 1.1 million molecular forms drawn from crystal structures. When they tested it on 5,075 ligands with multiple possible forms, it flagged 126 existing assignments as likely incorrect. Those assignments warrant review; the result does not mean the experimentally determined protein structures were wrong.