Hello, A change in allocation has been requested: Requester: wjiang (wei jiang) Project: estrogen_namd Title: Long-scale conformational activation in estrogen receptor with NAMD on GPU/CPU heterogeneous architecture Description: Wei Jiang, Sunhwan Jo, Sichun Yang Leadership Computing Facility, Argonne National Laboratory Case Western Reserve University, School of Medicine Projection Description Figure 2 Simulations show large-scale dynamics near the H12 region. Four conformations (in yellow) obtained from all-atom NAMD simulations show the loop region near H12 is highly flexible. Figure 1 Two states of the LBD. The C-terminal helix (H12) adopts a completely different orientation. Overall. The ligand-binding domain (LBD) of an estrogen receptor undergoes a large conformational switching from an inactive to active state in response to hormone stimuli (see Figure 1). The inactive conformation is preferred by therapeutic inhibitors such as tamoxifen, where a C-terminal helix H12 (marked in red and blue in Figure 1) blocks the receptor from recruiting co-activators. Once activated by estrogen binding, H12 is switched to a completely different position (marked in red) that covers up the hormone-binding pocket in the active conformation. These two well-defined conformations—active and inactive—have been revealed extensively by crystallographic studies. Very recently, a novel D538G mutant has been identified to be active in advanced breast cancer tumors. Despite being the key molecular target of breast cancer therapeutics, the LBD is a remarkable example of conformational switching, where molecular dynamics of LDB, especially the region near H12, from an inactive to active conformation remains unclear. The first goal of this proposal is aimed at using high-performance molecular dynamics simulations to quantify the motion near the H12 region for structure-based drug discovery. The second goal is finding novel drug molecules against the receptor domain via accurate free energy calculations, based on those important sub-states identified in the first goal. Preliminary findings on receptor dynamics simulations. Our published coarse-grained molecular dynamics shows such a transition could take at least in the order of 100 ns (Huang et al, JCTC, 10 (8) 2897–2900, 2014). While this information is highly informative, a key question remains, i.e., whether the mutation affects the active state only or sculpts the entire landscape, including both inactive and active states. More recently, supported by a LCRC award, we have begun to answer this question by simulating the receptor dynamics. Figure 2 shows our preliminary findings about a very diverse set of loop conformations near the H12, suggesting that the system is highly flexible and requires large-scale simulations for sufficient sampling. Note that the simulations were based on the active LBD conformations only. A full H12 transition from inactive to active can be quite challenging for atomic-level simulations using the standard CPU-based MD simulations, and obtaining such a com plete set of loop/H12 conformations is highly promising for structure-based drug design. To obtain better sampling and long-time MD trajectories, we propose to extend this study by using the GPU version (Nvidia CUDA) NAMD package on the Blues bigmem GPU/CPU nodes. NAMD is the only one MD code that can run at strong scaling on massively distributed GPU/CPU platform such as Cray XT7. In this proposal, in addition to the brutal-force GPU/CPU work, we also explore AVX256 instruction set to optimize force kernel of NAMD for state-of-the-art Intel architecture. A fully vectorized force kernel with AVX256 also provides a better reasonable CPU performance reference for GPU. In this proposal we apply a total 50,000 core hours, including both long production run on GPU/CPU nodes and code development with AVX256. Figure 3 Accurate docking and binding affinity calculations. A set of four well-studied ligands were used in calculations. Their computed free energies are highly correlated with experimental measurements. Preliminary findings on docking/binding affinity calculations. Identifying potent ligand-receptor binding is a primary objective of small molecule screening with regards to environmental safety and biological efficacy. During the past decades, computational docking and binding affinity calculations have benefited from improved force fields and sampling algorithms, as well as the advent of highly efficient parallel computing. Such developments have enabled free energy calculations to yield accurate binding affinity predictions, especially via a well-known method of free energy perturbation (FEP). Here, we will apply a docking-FEP protocol on receptor conformations identified from NAMD simulations to achieve the accuracy and reliability to influence lead optimization in a drug discovery setting. This computational method conducts a two-tier evaluation of all ligands: a high-speed docking evaluation by robust binding mode predictions and a high-performance binding affinity evalu ation via accurate free energy calculations. To demonstrate how the calculations perform, we examined several ligands with four well-known experimental binding affinities to estrogen receptor (ER), supported by a recent LCRC award. Based on this set of ligands, we docked them into the hormone-binding site and applied free energy calculations to each docked ligand-receptor mode. We found a high-level accuracy across a broad range of free energies, with a remarkable agreement to their experimental binding data available in the literature. Figure 3 shows that we achieved a high level of accuracy by comparing the computed and experiential binding affinities of these ligands to the receptor. These highly consistent results from this pilot study are part of primary studies for this renewal for targeting a diverse set of receptor conformations that will be determined by large-scale NAMD simulations. 2015 allocation request: 100000 Q3: 50000 Q4: 50000 Justification: The request for Q3 (July 1st to Sept 30) and Q4 (Oct 1st to Dec 31st) is based on our core-hour usage and achievement during Apr 1 and June 30rd. Current: undetermined amount Justification: NAMD is the only one MD code that can run at strong scaling on massively distributed GPU/CPU platform such as Cray XT7. Requested: 100000 A specific reason has been given: High throughput drug screening based on the structural information obtained with previous MD/REST2 simulation on Blues/Fusion. This needs to be approved and the final allocation amount decided upon. Thank You, The LCRC Accounts System