《生命科学》 2025, 37(7): 894-903

蛋白质动态结构解析研究进展

蔡幸初¹ , 李兆达^2,* , 张冬妮² , 肖凡玥² , 贺　明^2,*

¹湖北思朗万维计算装备制造有限公司，孝感 432012 ²长江3D科学计算中心，孝感 432015

摘　要：

蛋白质结构解析是理解生命机制与药物研发的核心。本文系统回顾了人工智能(artificial intelligence, AI) 驱动下的蛋白质结构预测的进展、分子动力学(molecular dynamics, MD) 模拟在揭示动态构象与分子互作中的发展，以及高性能计算( 如专用科学计算芯片MHPC512) 对长时程模拟发挥的作用。研究表明，AI预测模型已实现从静态单体到多分子复合体( 蛋白质- 核酸- 配体) 的高精度解析，但在动态行为模拟和侧链预测上仍存局限性。经典MD 模拟通过改进增强采样策略，已能捕捉微秒级蛋白质构象变化，成为揭示药物结合机制与蛋白质功能动态的关键工具。高性能计算( 如Anton 系列、国产MHPC512) 已经突破算力瓶颈，使百万原子体系毫秒级模拟成为可能，为构建蛋白质动态结构库奠定基础。未来需深度融合“AI 预测初筛+MD 动态精修+ 实验验证”的范式，推动靶向动态构象的创新药物设计。大规模的MD 蛋白质动态结构数据库，将在未来的科研和工业端发挥巨大的应用价值。

通讯作者：李兆达 , Email:zhaoda.li@smartlogictech.com 贺　明 , Email:ming.he@smartlogictech.com

Research progress of protein dynamic structure analysis

CAI Xing-Chu¹ , LI Zhao-Da^2,* , ZHANG Dong-Ni² , XIAO Fan-Yue² , HE Ming^2,*

¹Hubei Universal Logic Calculating Equipment Manufacturing Co., Ltd, Xiaogan 432012 ²Yangtze River 3D Scientific Computing Center, Xiaogan 432015, China

Abstract:

Protein structure determination serves as the cornerstone for understanding biological mechanisms and drug discovery. This review systematically examines the revolutionary advances in artificial intelligence (AI)-driven protein structure prediction, the indispensable role of molecular dynamics (MD) simulations in elucidating dynamic conformations and molecular interactions, and the breakthrough support provided by high-performance computing (HPC), exemplified by specialized supercomputing chips like MHPC512, for enabling long-timescale simulations. Key findings indicate: AI prediction models have achieved high-precision resolution from static monomers to multimolecular complexes (protein-nucleic acid-ligand), yet remain limited in simulating dynamic behavior and accurately predicting side-chain conformations. Classical MD simulations, enhanced by refined sampling strategies, can now capture microsecond-scale protein conformational changes, establishing themselves as critical tools for revealing drug binding mechanisms and functional protein dynamics. High-performance computing breakthroughs (e.g., Anton series, domestic MHPC512) overcome computational bottlenecks, enabling millisecond-scale simulations of million-atom systems and laying the foundation for constructing dynamic protein structure libraries. Future progress necessitates the deep integration of an "AI Initial Screening + MD Dynamic Refinement + Experimental Validation" paradigm to propel the design of innovative drugs targeting dynamic conformations. Crucially, large-scale MD protein dynamic structure databases are expected to deliver immense value in both research and industrial applications.

Communication Author：LI Zhao-Da , Email:zhaoda.li@smartlogictech.com HE Ming , Email:ming.he@smartlogictech.com