ProtMD: Pioneering AI Model for Dynamic Protein Conformation and Drug-Target Affinity Prediction

Westlake University announced on the 8th that the team of Li Ziqing, the AI Chair Professor at the university, in collaboration with Xiamen University and MindRank, pioneered the development of an AI model called ProtMD, which can characterize protein conformational changes and predict affinity.

This is the first artificial intelligence method to attempt parsing protein dynamic conformations, which can assist medicinal chemistry experts in more accurately screening high-activity small molecules, thereby accelerating preclinical drug development. The relevant research findings were published in the journal *Advanced Science*.

Li Ziqing introduced that the "AlphaFold2," previously developed by a Google-owned company, can accurately predict the three-dimensional structure of proteins using artificial intelligence, which has had a tremendous impact on structural biology, drug design, and even the entire scientific community. However, "AlphaFold2" can only predict the static structure of proteins at a single moment and has not yet solved the prediction of dynamic changes in protein structures.

The AI model developed this time by Li Ziqing's team can predict the process of structural changes in target proteins after drug molecules bind to them (flexible docking) within the body, infer the stability of the drug-target protein binding, and predict drug functionality, thereby improving the accuracy and efficiency of AI-driven drug design.

The research team first selected dozens of representative protein structures from 57,651 human protein structures for molecular dynamics simulations to obtain the spatial motion trajectories of the proteins and establish models of their dynamic conformations. During the pre-training phase, the team required the model to predict the next conformation of a protein based on its previous conformation and simultaneously trained the model to sort protein sequences at different times, enabling it to reorder protein conformations with shuffled sequences. Experiments show that this AI model, even in its lightweight version, has surpassed existing state-of-the-art models in drug-protein affinity prediction tasks.

"Predicting the dynamic changes of protein structures is of great significance for understanding life processes and developing new drugs," said Li Ziqing. Especially in AI drug design, predicting the dynamic structural changes after drug molecules bind to target proteins, and evaluating drug-target binding affinity and drug efficacy, are important approaches to improve the accuracy and efficiency of AI drug screening.