Stochastic road-map simulation is a tool designed to efficiently estimate ensemble properties and other key properties of molecular motion. Unlike previous approaches, it does not perform individual simulation runs. Instead, it pre-computes a road-map by sampling the conformational space at random and uses tools from Markov chain theory to computer properties over the many molecular pathways encoded in the road-map. We have adapted stochastic road-map simulation techniques to non-uniform probabilistic road-maps and developed several non-uniform sampling strategies, e.g., a Gaussian strategy that places more samples in areas of the conformational space where the energy undergoes quick variations.

Concurrently, we have incorporated new protein representations and energy functions (Go-model, C-based representation) in stochastic road-map simulation, as well as additional computational tools such as the nearest-neighbor algorithm of Lotan and Schwarzer. We have obtained pfold (probability-of-folding parameter) results for a 42-DoF model of the beta hairpin protein that agrees well with Monte Carlo simulation. A modularized version of the stochastic road-map simulation software is in preparation and will soon be released on the web.