Molecular: Markov state models for simulation

States and transitions during protein folding (source: Voelz et al. JACS 2010, 132, p 1526)

Molecular dynamics simulation has evolved from a severely limited esoteric method into a cornerstone of many fields, in particular structural biology where it is now just as established NMR or X-ray crystallography. To achieve high performance, the simulations are typically run on massively parallel computers using domain decomposition of calculations, and for large enough systems (hundreds of millions of particles) this can scale to very large machines. However, a central challenge for the applications is that most molecules of biological interest (e.g. proteins) only require maybe 50,000-250,000 atoms. Since we would still like to reach simulation timescales many orders of magnitude larger than today this would lead to unreasonable requirements for strong scalability.

To get around these limitations, we are developing a new generation of parallel adaptive molecular dynamics methods that rely on Markov State Models. By running a large number of independent (but parallel) simulations for a system it is possible to cluster conformations and identify common states visited in several simulations, and construct a transition probability matrix between states. This makes it possible to enhance sampling by increasing the weight of the least visited states, but the most important feature is that it becomes possible to study slow processes on timescales much longer than the individual simulations as a composition of smaller transitions.

Our central research topics in this project concern methods to assess convergence and how the MSM approach can be extended from exhaustive sampling of small molecules into targeted studies of specific conformational transitions, in particular for membrane proteins. This is a multi-disciplinary project with branches both into numerical analysis, techniques for utilizing extremely parallel hardware (including the K computer) efficiently, and not least data handling.

Contact person: Erik Lindahl, KTH/SU
Other researchers involved: Anders Gabrielsson, Iman Pouya, Sander Pronk, KTH