¡New scientific publication!
Today we are celebrating great news: our researcher Ana María Fernández Escamilla has just published a new study analyzing the performance of various deep learning tools and first-principles-based methods for protein redesign, with a focus on their therapeutic aplicability.
Title of the article: Artificial intelligence and first-principle methods in protein redesign: A marriage of convenience?
DOI: In this work:
The ability of diferent approaches —from force fields to inverse folding tools like — is evaluated for predicting the effect of multiple simultaneous mutations.
TriCombine, a new tool, is presented. It analyzed residue triplets in proteins to guide the selection of promising mutants.
Predictors are validated using a collection of 36 mutants, including 11 cristal structures, and a laarge-scale analysis of over 300.000 variants in both natural and the novo domains.
The article concludes that combining AI-based modeling with physics-based scoring functions (such as FoldX) delivers the most realible results, specially when redesigning proteins with multiple mutations.
The work highlights the importance of hybrid strategies in designing robust proteins, particularly when dealing with de novo proteins or proteins without resolved crystal structures.
A big thank you to all the authors for this outstanding multidisciplinary effort!