Scientists have found a new way to predict how proteins change shape as they function, which is important for understanding how they work in living systems. Although recent artificial intelligence (AI) technology has made it possible to predict what proteins look like in their resting state, figuring out how they move is still a challenge because there isn’t enough direct data from protein movement experiments to train the neural networks.
From a new study published in the Proceedings of the National Academy of Sciences on August 20, Peter Wolynes of Rice University and his colleagues in China combined information about protein energy landscapes with deep learning techniques to predict these movements.
Their method improves on AlphaFold2 (AF2), a tool that predicts static protein structures, by teaching it to focus on “energetic frustration.” Proteins have evolved to minimize energetic conflict between their parts so that they can be guided toward their static structure. Where conflict persists, there is frustration.
“Starting from predicted static ground state structures, the new method generates alternative structures and routes for protein movement by first finding and then gradually improving the energetic frustration signatures in the input multiple sequence alignment sequences that encode the evolutionary development of the protein,” Wolynes said. the DR Bullard-Welch Foundation Professor of Science and co-author of the study.
The researchers tested their method on the protein adenylate kinase and found that its predicted movements matched experimental data. They also successfully predicted the functional movements of other proteins that change shape significantly.
“Predicting the three-dimensional structures and movements of proteins is an integral part of understanding their functions and designing new drugs,” Wolynes said.
The study also examined how AF2 works, showing that combining physical knowledge of the energy landscape with AI not only helps predict how proteins move, but also explains why the AI overpredicts structural integrity, which only affects the most stable structures leads.
The energy landscape theory, which Wolynes and his collaborators have been working with for the past decades, is a key part of this method, but recent AI codes have been trained to predict only the most stable protein structures and ignore the different shapes that proteins can take as they function . .
Energy landscape theory suggests that although evolution has shaped the energy landscape of proteins where they can fold into their optimal structures, deviations from a perfectly channelized landscape that otherwise directs folding, called local frustration, are essential for functional movements of proteins.
By pinpointing these frustrated regions, the researchers taught the AI to ignore these regions when driving its predictions, allowing the code to accurately predict alternative protein structures and functional movements.
Using a frustration analysis tool developed within the energy landscape framework, researchers identified frustrated and therefore flexible regions in proteins.
By then manipulating the evolutionary information in the aligned protein family sequences used by AlphaFold and corresponding to the frustration scores, the researchers taught the AI to recognize these frustrated regions, enabling accurate predictions of alternative structures and pathways between them, Wolynes said.
“This research underlines the importance of not forgetting or abandoning physics-based methods in the post-AlphaFold era, where the emphasis has been on agnostic learning based on experimental data without any theoretical input,” said Wolynes. “Integrating AI with biophysical insights will have a significant impact on future practical applications, including drug design, enzyme engineering and understanding disease mechanisms.”
Other authors include Xingyue Guana, Wei Wanga and Wenfei Lia from Nanjing University’s Department of Physics; Qian-Yuan Tang from the Department of Physics at Hong Kong Baptist University; Weitong Ren from the Wenzhou Key Laboratory of Biophysics of the University of Chinese Academy of Sciences; and Mingchen Chen at the Changping Laboratory in Beijing.
The work was supported by the National Natural Science Foundation of China; Wenzhou Institute, University of the Chinese Academy of Sciences; Hong Kong Research Grant Council; and the US National Science Foundation-funded Center for Theoretical Biological Physics at Rice.
