Protein dynamics are essential for understanding how proteins function and occur on a wide range of timescales. Most functional dynamics occur on far-reaching timescales that are intractable to brute force molecular dynamics; let alone quantitatively decipt these pivotal events. The above situtation  holds even to rarely accessible resources like that of the D.E. Shaw Anton computer. The present study aims at exploring the possibiity of overcoming the timescale limit via the combination of novel free energy sampling techniques and the commodity GPU hardware. Specifically, we implemented the high-order orthogonal space tempering technique in the new CHARMM DOMDEC-GPU routine via the CUDA programming laugauage; it is worth noting that this implementation introduces negligible computing overheads. In this study, we investigated two biologically pivotal processes, the DFG flipping of tyrosine kinases (the C-Src and Abl kinases), which was extensively explored through various enhanced sampling methods, and the activation of Heterotrimeric G protein through the opening of the α subunit domains, which was recently simulated through the Anton computer. Based on our strategy, for the first time, these long-timescale processes can repetitively occur in single continuous trajectories; as a result, we can confidently map the corresponding free energy landscapes. Our results not only confirm some known insights but also are able to meld the essential missing links that cannot be obtained from earlier advanced studies. For instance, the results from these single-GPU based simulations firmly show that the potential of mean force calculations from a national computing system indeed captured an inaccurate major free energy minimum and the results from the Anton simulations cannot access to key activation events. The success of our GPU-based studies showcases the power of the high-order orthogonal space tempering technique; more importantly, it suggests that combining state-of-the-art enhanced sampling methods with the GPU computing environment is a viable path to routinely tract long-timescale protein functional motions that has been a long-pursued challenge in the field of  biomolecular simulation.

Routinely tracting long-timescale protein motions via GPU-based orthogonal space sampling molecular dynamics

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.