论文部分内容阅读
Markov State Model(MSM)becomes a popular approach in investigating conformational dynamics of complex systems.But in many cases,one often needs a large number of states(e.g.1000 to 10000 microstates)to construct a Markovian model,and any attempt to further reduce this number will render the model non-Markovian unless a much longer lagtime is used.On one hand,models with too many states will hinder easy comprehension of biological insights,and kinetic lumping is often required to reduce the number of states(i.e.macrostates).On the other hand,when the number of states is small,a much longer lagtime is needed for the model to be Markovian,while the lagtime is limited by the length of molecular dynamics simulations.In this talk,I will introduce a new method to obtain kinetic lumped macrostate MSMs that can reproduce the slowest dynamics of the system(from microstate MSM)but with very short lagtime.To achieve this,we apply the projection operator framework to backward project coarse-grained macrostate transition modes(or eigenmodes)to slowest transition modes of a detailed microstate Markov chain,so that we can compare them against those from the original microstate MSM.Once we identify a kinetic lumping with matching eigenmodes,we can utilize the lumping and reconstruct a macrostate MSM that can faithfully reproduce the slowest kinetics of the system.This will allow us to construct good macrostate MSMs with very short lagtime.Finally,I will show our initial attempts to apply this new method to identify best reaction coordinates to investigate protein conformational changes.