Drug firms’ secret data supercharge AI protein models

An AI system trained on over 20,000 proprietary protein structures from pharmaceutical companies significantly outperforms AlphaFold-like models built solely on publicly available data. The results demonstrate that private structural datasets held by drug makers contain information not captured by public repositories, enabling more accurate prediction of protein conformations critical to drug design. The advance has broad implications for accelerating target-based drug discovery across therapeutic areas, including oncology, where precise protein modeling underpins the development of small-molecule inhibitors and biologics.