On the Reliability of AI Methods in Drug Discovery: Evaluation of Boltz-2 for Structure and Binding Affinity Prediction

A comprehensive evaluation of Boltz-2, an AI-based drug discovery tool, reveals significant limitations in predicting protein-ligand binding structures and affinities. The study found only weak correlations with physics-based methods and concluded that while useful for initial screening, Boltz-2 lacks the precision required for reliable drug lead identification.

• →Boltz-2 showed significant structural prediction variations, indicating multiple protein conformations rather than converged poses.
• →Binding affinity predictions exhibited only weak to moderate correlations with established physics-based methods.
• →Analysis of top 100 compounds showed no significant correlation between Boltz-2 predictions and binding free energies.
• →The study highlights that no AI-discovered drugs have yet received regulatory approval despite industry hype.
• →Physics-based methods remain necessary for reliable refinement of AI-derived drug discovery models.