845-Pos Board B645 Improved Library-Based Monte Carlo, Applied to Multi-Level Sampling Ying Ding, Artem B. Mamonov, Daniel M. Zuckerman. We describe further advances in our previously developed library-based Monte Carlo (LBMC) simulation. LBMC is a memory-intensive molecular simulation strategy which uses pre-calculated configurations of molecular fragments, such as amino acids or side-chains. We showed previously that LBMC using whole-residue fragments could lead to dramatic efficiency gains for sampling flexible peptides; however, such fragments are not ideal for simulating folded systems because trial moves are highly non-local. We therefore have developed a new implementation of LBMC, by separating the backbone from side-chain libraries, which allows for exactly local moves perturbing only a small part of a protein. Because the positions of all atoms are carried in memory at minimal run-time cost, LBMC readily permits the use of hybrid protein models mixing full and reduced interactions. Hybrid models are useful in themselves-eg, to simulate an atomistic binding site allosterically coupled to a reduced description of the remainder of the protein. Here, however, we explore the use of hybrid models in multi-level equilibrium sampling algorithms based on Hamiltonian and resolution exchange. Higher ladder levels are built to allow cheaper energy calls and smoother potential energy landscapes to facilitate sampling of low levels.