(527f) Large-Scale All-Atom Conformational Sampling of Proteins Using Temperature-Accelerated Molecular Dynamics

We present a new method for large-scale conformational sampling of proteins in all-atom explicit-solvent MD simulations based on the theory of Temperature Accelerated Molecular Dynamics (TAMD) [Abrams and Vanden-Eijnden, PNAS 107:4961 (2010)]. To use TAMD, one augments a standard MD simulation system with a relatively small number of ?slow? variables, each of which is tethered by a stiff harmonic spring to a counterpart collective variable which is a function of atomic configuration. These slow variables evolve in lock-step with the atomic variables but obey overdamped dynamics with high friction and high (fictitious) temperature. The high friction guarantees that the forces ?felt? by the slow variables from the atomic system approximate negative gradients in the free energy associated with the set of collective variables. The slow variables evolve at high temperature on the free energy surface computed at the physical temperature, and therefore can drive the system easily over moderate free energy barriers. We show that by adopting collective variables that are cartesian coordinates of discrete blocks of protein structure, or ?subdomains?, TAMD can accelerate conformational sampling in large multidomain proteins and succeeds at predicting crystallographically observed conformations. In particular, we show how TAMD can induce the t-to-r? transition in the GroEL subunit and the CD4-bound to F105-bound transition in the HIV-1 gp120 core. TAMD-based conformational sampling should prove useful in generating all-atom models of so-far unobserved conformations of large proteins, thereby potentially aiding in the characterization of new drug design targets.