Multiscale modeling of molecular systems with data-driven collective variables

BEHROOZ HASHEMIAN; DANIEL MILLÁN; MARINO ARROYO

Providence, RI

Congreso; 50th Annual Technical Meeting of the Society of the Engineering Science (SES 2013); 2013

Society of Engineering Science

When modeling complex molecular systems, such as proteins, there is often an overwhelming gap between the computable time-scales and those required for thermodynamical sampling, particularly in the presence of metastability. Consequently, comparison with experiments is difficult, and plagued with uncertainties related to insufficient sampling. Free energy calculation methods can alleviate such challenges, but require a good set of collective variables (CVs) to be effective. Identifying such a coarse model for the system is far from obvious for complex systems. Here, we present a general method based on statistical learning to identify such CVs from available computational or experimental ensembles, and integrate them in enhanced sampling methods. The method is minimally invasive with respect to the underlying molecular dynamics code. We exercise the method with small peptides and membrane proteins.