Can pharmaceutical companies improve AI predictions while keeping their research data confidential?
Silvio Tosatto, head of the Biocomputing UP lab at the University of Padua, is among the experts featured in a recent Le Scienze article by Chiara Sabelli exploring this question.
The article examines an industry collaboration using federated learning to develop AI models for predicting protein–molecule structures. This approach allows companies to contribute to model training without directly sharing their proprietary datasets.
Tosatto’s contribution focuses on the importance of rigorous, independent evaluation. He points to CASP, an international assessment of structure prediction methods, as a possible framework for testing the reported advances impartially.
For researchers considering these tools, such evaluation is essential to understanding how improvements in prediction translate into practical scientific value.
Read the original article: Chiara Sabelli, “Cinque aziende farmaceutiche hanno unito i loro dati (in segreto) migliorando i modelli di IA che prevedono molecole di interesse terapeutico”, Le Scienze, 22 September 2026, GEDI Periodici e Servizi S.p.A.
Full article in Italian; subscription required.