Scientists at Skoltech Bio have evaluated AlphaFold, the AI program renowned for predicting protein structures, on its ability to assess the impact of single mutations on protein stability. The investigation revealed that AlphaFold’s predictions often contradicted established experimental data, undermining claims that it had achieved mastery over protein physics and could address broader challenges in structural bioinformatics. Although AlphaFold successfully resolved the long-standing issue of predicting protein structures—where it identifies 3D shapes based on amino acid sequences—its limitations became evident when tasked with predicting mutation effects. The team emphasized that even with impressive capabilities, current AI technologies do not yet serve as a panacea for all scientific questions in the field. Notably, the study involved high school students who contributed to processing mutation data and exploring the AI’s functionalities. Their findings indicated that AI has not accurately internalized the underlying physics of protein behavior, as evidenced by specific inconsistencies in mutation prediction. Ultimately, while AlphaFold represents a significant advancement, the research reinforces the notion that major challenges remain in structural bioinformatics, where human insight and further scientific exploration are essential.
