Genome-wide-scale prediction of compound-protein interactions using foundation and language models based on three-dimensional structures of compounds and proteins.

DESTIN predicts genome-wide compound-protein interactions with high accuracy by utilizing 3D structural information, achieving strong results even for membrane proteins.

• Why it matters: Accurate identification of CPIs is vital for drug discovery, but existing methods lack the ability to incorporate detailed geometric binding site information, limiting their effectiveness.
• What they did: The approach involved developing a machine learning framework that constructs spatially informed compound features from quantum-chemical 3D conformations and leverages sequence-based structure models for proteins, including those without experimental 3D data, using 3D-aware feature vectors.
• The result: DESTIN demonstrated robust predictive performance across benchmark tests, especially for complex membrane proteins, enabling more reliable CPI predictions and advancing pharmaceutical research capabilities.