- IsoDDE significantly outperforms AlphaFold 3 in protein-ligand structure prediction, more than doubling accuracy on challenging benchmarks.
- The model predicts small molecule binding affinities with higher accuracy than physics-based methods like FEP, while being much faster and cheaper.
- IsoDDE can accurately identify novel binding pockets on proteins using only amino acid sequences, even for cryptic or allosteric sites.
- It shows exceptional performance in antibody-antigen interface modeling, especially for the variable CDR-H3 loop, enabling de novo antibody design.
- The system generalizes well to unseen biomolecular systems, handling complex events like induced fits and cryptic pocket opening.
- IsoDDE provides scalable, unified computational drug design, bridging structure prediction and real-world drug discovery applications.