Researchers have found a new way to teach artificial intelligence (AI) about the frustrations involved in protein folding, ultimately aiding in our understanding of how proteins function within living systems. While AI technology has advanced in predicting static protein structures, deciphering how proteins move remains a challenge due to limited experimental data on protein motions.
In a recent study published in the Proceedings of the National Academy of Sciences, scientists from Rice University and China combined information on protein energy landscapes with deep-learning techniques to predict these movements effectively. By focusing on energetic frustration, a concept where proteins evolve to minimize conflicts between their components, the researchers enhanced AlphaFold2 (AF2) to predict protein motions accurately.
Peter Wolynes, co-author of the study, highlighted the significance of their method in generating alternative protein structures and pathways based on energetic frustration features. The researchers successfully applied this method to analyze the protein adenylate kinase and predicted functional movements in other proteins with shape variations.
Understanding the three-dimensional structures and motions of proteins is vital for drug design and comprehending their functions. By integrating physical knowledge of energy landscapes with AI, the study not only predicts protein movements but also explains the tendency of AI to overpredict structural integrity.
The incorporation of the energy landscape theory, coupled with AI, proves instrumental in accurately predicting protein structures and pathways essential for protein functional movements. By identifying and addressing frustrated regions within proteins, the researchers taught the AI to ignore these regions in its predictions, leading to more precise outcomes.
This research underscores the importance of combining AI technology with biophysical insights for practical applications in drug design, enzyme engineering, and disease mechanism understanding. The findings emphasize the value of not disregarding physics-based methods in the post-AlphaFold era, heralding a new era of AI integration in protein research for diverse applications.
