Exploring Protein Conformational Changes Using a Large-Scale Biophysical Sampling Augmented Deep Learning Strategy.

Area of Science:

• Biophysics
• Computational Biology
• Structural Biology

Background:

• Deep learning models have advanced static protein structure prediction.
• Predicting dynamic protein conformational changes remains a significant challenge due to limited training data.
• Understanding protein dynamics is crucial for deciphering biological functions.

Purpose of the Study:

• To develop a general deep learning model for predicting protein conformational transition pathways.
• To address the scarcity of training data for protein dynamics by creating a large-scale simulation database.
• To validate the model's predictive power and its application in biological discovery.

Main Methods:

• Generated a large-scale database by combining molecular dynamics simulations with enhanced sampling methods for 2635 proteins.
• Simulated conformational changes between two known stable states for each protein.
• Developed and applied a general deep learning model to predict protein transition pathways.

Main Results:

• The deep learning model demonstrated robustness across proteins of varying lengths and diverse conformational changes.
• Predictions showed strong agreement with experimental data in multiple test systems.
• The model successfully identified a novel allosteric regulation in human β-cardiac myosin.

Conclusions:

• The developed deep learning model is effective for predicting protein conformational changes.
• The approach of combining simulations with deep learning provides a powerful tool for studying protein dynamics.
• This work advances our understanding of protein conformational transitions and their biological implications.