This proposal is to develop MELD, a computational Bayesian accelerator that ?melds? together molecular dynamics simulations with external knowledge. It is novel in harnessing information that has not been usable before ? because it is too sparse, noisy, ambiguous, combinatoric, or too corrupted for traditional approaches. In contrast to the high-certainty restraints traditionally used in MD simulations, MELD leverages a much broader range of real-world high-uncertainty restraints. The first specific aim is to incorporate such information in protein structure determination, in several collaboration projects with experimentalists who perform solution x-ray scattering, ESR, and high-throughput alanine scanning structures of peptide protein complexes.
The second aim i s to also harness information about processes, trajectories, and dynamic routes to speed the identification of protein states. MELD promises to extend physics-based simulations for determining larger protein structures, for folding larger proteins, for binding more flexible ligands, and for exploring larger mechanistic actions, than current MD simulation methods can handle.
Biomedical research and pharmaceutics depend on detailed understanding of the structures and motions of proteins. Molecular dynamics simulations provide the most detailed descriptions possible, however they cannot yet describe average to large sized protein structures or motions within a reasonable time frame. We propose to develop a new physics-based computational accelerator for Molecular Dynamics, called MELD, which incorporates many types of relevant external information that was too vague and difficult to compute to have been practically useful before.