TL;DR
Isomorphic Labs releases IsoDDE, a drug design system that more than doubles AlphaFold 3's accuracy on generalization benchmarks and surpasses physics-based methods for binding affinity prediction.
Key Points
- IsoDDE demonstrates 2x+ accuracy improvement over AlphaFold 3 on 'Runs N Poses' generalization benchmark for novel protein-ligand structures
- Outperforms AlphaFold 3 by 2.3x and Boltz-2 by 19.8x on antibody-antigen interface prediction in high-fidelity regime
- Surpasses physics-based FEP methods on binding affinity prediction while running in seconds instead of hours/days, without requiring experimental crystal structures
- Identifies cryptic and allosteric binding pockets from protein sequences alone; recapitulated recent cereblon cryptic pocket discovery using only amino acid sequence input
Why It Matters
This advances computational drug discovery from structure prediction to actionable binding predictions and pocket identification—compressing months of experimental work into seconds. For researchers and biotech teams, IsoDDE enables faster exploration of novel therapeutic targets and mechanisms that were previously inaccessible, reducing both computational cost and time-to-lead optimization.
Source: www.isomorphiclabs.com