A Robust Framework for Adaptive Multiscale Modeling of Biopolymers Using Highly Parallelizable Methods

For many biopolymers (RNA, DNA, enzymes and proteins) the nature of the molecules interaction within the cell has been determined to be highly a function of its conformational structure. Understanding how to influence and control this structure thus is of critical importance if one wishes to manipulate the intercellular processes of which these biopolymers play such a central role. In molecular dynamics (MD) simulations, a fully atomistic model represents the system at the finest scale and as such captures all the dynamics of the system. If the simulation is permitted to run sufficiently long important emergent behaviors can develop and show themselves. Such MD simulations represent a direct applications of Newton’s Laws of Motion to the individual atoms in the system, and are conceptually the easiest to implement. An advantage of this procedure is that the simulation yields important information not only about the intermediate states and the mechanisms which produced them, but also provides the rates at which these processes occur. These intermediate conformational states have repeatedly been implicated in many known biological function [1], [2]. Unfortunately, this albeit correct, but naive approach quickly leads to intractable models and prohibitive computational expense when applied to systems involving many atoms. As a result, researcher often grossly over simplify the system move to non-deterministic methods such as Monte Carlo, which effectively remove dynamics from the system, or use undesirably gross model simplification. Because of these forward dynamics performance difficulties, potentially important mechanisms governing biopolymer structure have not been adequately explored and/or identified. The methods and algorithms described in this paper are intended to extend the capabilities of the simulation techniques for such systems so that the forward dynamics can better predict the non-equilibrium behavior of these systems, thus complementing Monte Carlo, while retaining much useful intermediate process and temporal information.