Markov State Model approaches for folding, binding and design

点击翻译按钮获取中文摘要

Project Summary Understanding the conformational dynamics of proteins and their binding partners is crucial to predicting and designing their function. Molecular simulations are suitable for this task, but remain challenging for ligand binding systems where dissociation occurs on very slow time scales. We are developing new Markov state model (MSM) approaches, which describe conformational dynamics as a network of transitions between metastable states, to address this challenge. Multi-scale Markov models (MEMMs) offer a robust framework for building variationally optimal models of dynamics on long time scales, from ensembles of short trajectories sampled in biased thermodynamic ensembles, to predict ligand binding affinities, rates and mechanisms. During the coronavirus pandemic, our group used the distributed computing platform Folding@home (FAH) to perform virtual screening of SARS-CoV-2 main protease inhibitors by utilizing expanded-ensemble (EE) simulations, in which multiple alchemical intermediates can be sampled in a single simulation, to estimate binding free energies. This has inspired us to combine EE and MSM methods that can leverage the power of FAH to make fundamental advances in virtual screening and molecular design, in three specific aims: Our first aim is to improve EE methods for computing ligand binding free energies. In collaboration with the Shirts Lab, we seek to understand and ameliorate convergence issues, and explore and unify related approaches. We will investigate how well EE estimates of free energies of mutations can be used with MEMMs to predict changes in protein folding stability and rates. Finally, we will work with the Karanicolas Lab to determine the extent to which EE-calculated ABFEs on FAH can be used alongside advanced machine learning classifiers to discover both active and potent inhibitors from structure-based virtual screening studies. Our second aim is to develop a combined metadynamics (metaD) + MEMM approach for modeling binding reactions. We will develop and test two different strategies in which metaD is used to derive negative potentials of mean force along binding reaction coordinates that can be used as bias potentials for constructing multi- ensemble Markov models (MEMMs) of ligand binding. We will test these methods in toy binding systems, and small ligands of L99A lysozyme. Finally, we will apply metaD+MEMMs to predict affinities, rates and mechanisms of the macrolide natural product carolacton binding to FolD and its known drug-resistant mutants. Our third aim is to examine the extent to which solution-state preorganization determines binding affinity, and whether simulation-based modeling can use this idea quantitatively for computational design. For a corpus of 105 cyclic peptides with published affinities, EE+MSM approaches will test the validity of a two-step conformational selection model. The results of this work will guide the design, testing and optimization of cyclic peptide binders to disrupt dimerization of the tumor suppressor PTEN, a collaboration with the Rongsheng Wang Lab at Temple, to find new diagnostics/therapeutics for cancer metastasis.