Interpretable Deep Supramolecular Language Modeling for Cocrystal Design: PCA-Augmented DeepCocrystal Screening of Picolinamide, Nicotinamide, and Isonicotinamide Systems

Description

Bahadır Bozkurt and four colleagues add a lens to an AI system that predicts which pairs will form drug cocrystals. Cocrystals can fix a drug's solubility or stability, but choosing a partner molecule is guesswork. The underlying DeepCocrystal model ranks candidate coformers for three related drug-like molecules. The new layer applies principal component analysis to the feature space without changing any prediction, to expose what drives them.

Three drivers dominated matching: hydrogen-bond matching, polarity matching, and aromatic surface contact. A simple acidity-based filter weeded out pairs likely to form salts instead of cocrystals. Of 137 pairs left positive or uncertain, the filter kept 46 as high-probability cocrystal candidates and 39 as uncertain with salt excluded. Another 34 landed in a salt-cocrystal gray zone, and 7 were flagged as salt risks. The authors pitch the combined approach as transparent, letting chemists prioritize coformers with reasons attached.

Catalog record

Source: https://europepmc.org/article/MED/PMC13610686