Machine-learning Assisted Molecular Simulations with Applications to Biological Systems
Speaker
Time
Venue
Online
Tencent
Abstract
Recently, molecular simulations have benefited greatly from the development and subsequent application of deep-learning methods. In this talk, we will discuss how machine-learning methods can be combined with enhanced sampling techniques to speed up molecular dynamic simulations. With these methods, one can perform efficient mechanistic studies at the atomic level for slow processes such as ice-water phase transition and chemical reactions in condensed phases. We will also discuss how deep molecular models can be constructed which allows investigation of chemical reactions at the quantum mechanics level but molecular mechanics computation cost. We have implemented these methods in our home-made molecular simulation package, SPONGE. The MD software was rewritten using MindSpore, to make it highly compatible with the machine learning platform. We will also discuss about our recent effort on learning from Alphafold2 and reproducing protein structure prediction under Mindspore and the MD simulaton package SPONGE. Through these efforts, we try to generate a comprehensive package of structure prediction, molecule and sequence generation, structure evaluation, and dynamics simulation. We will then discuss the possible applications of these methods in physical, chemical and biological problems.
Bio
高毅勤，1972 年出生，1993 年本科毕业于四川大学化学系，1996 年在中科院化学所获得硕士学位，2001 年获得加州理工学院博士学位。2001 年－ 2004 年在加州理工学院和哈佛大学做博士后研究。2004 年 -2010 年在美国德克萨斯农工大学（Texas A&M University）化学系任助理教授；2010 年起任北京大学化学与分子工程学院教授，2013 年起同时担任北京大学生物医学前沿创新中心研究员。主要从事生物物理化学/ 理论化学方面的基础研究。现任北京大学理学部副主任，JCTC杂志副主编，中国化学会副秘书长。
