Resources

Automated Collective Variable Discovery Protocol

Our protocol addresses the challenge of sampling transitions between long-lived metastable states in biomolecules by automating the detection of collective variables (CVs) for enhanced sampling in molecular dynamics (MD) simulations. Traditional CV selection relies heavily on intuition and prior knowledge, which can introduce bias and limit mechanistic insights. We circumvent this by utilizing machine learning algorithms to identify CVs automatically. For more details, please refer to the following publication: Oh M, Rosa M, Xie H, Khelashvili G. Automated collective variable discovery for MFSD2A transporter from molecular dynamics simulations. Biophysical Journal 2024 123(17):2934-55 PMID 38932456

Deep Neural Network-based machine learning analysis of molecular dynamic simulations for data classification

Our machine learning protocol transforms function-related, construct/state-specific differences encoded in molecular dynamics trajectories into visual representations recognizable by deep learning object recognition technology. The method then performs classification tasks with high accuracy and enables identification of molecular features of the molecular system that are major determinants for distinct conformations. For more details, please refer to the following publication: Oh M, Rosa M, Xie H, Khelashvili G. Automated collective variable discovery for MFSD2A transporter from molecular dynamics simulations. Biophysical Journal 2024 123(17):2934-55 PMID 38932456

N-body Information Theory (NbIT) based analysis of allosteric mechanisms from molecular dynamics simulations

N-body Information Theory (NbIT) uses measures of configurational entropy estimated from molecular dynamics simulations of a biomolecular system to identify microdomains and individual residues that act as (i)-channels for long-distance information sharing between functional sites, and (ii)-coordinators that regulate dynamics within functional sites. The formalism is based on the premise that, in order for a set of N structural components to propagate information effectively, they must achieve significant N-body mutual information. The NbIT thus uses the quasi-harmonic N-body mutual information to extract higher-order correlated motions in the atomic fluctuations present in equilibrium molecular dynamics simulations. For more details, please refer to the following publication: Khelashvili G, Kots E, Cheng X, Levine MV, Weinstein H. The allosteric mechanism leading to an open-groove lipid conductive state of the TMEM16F scramblase. Communications Biology 2022 5(1):990 PMID 36123525

Thermodynamic Coupling Function (TCF) calculation for quantifying allosteric mechanisms from molecular dynamics simulations

The Thermodynamic Coupling Function (TCF) provides a quantitative description of how particular states, or transitions between them in molecular dynamics simulations, are favored or opposed by allosteric coupling. The quantitative formalism estimates the contribution of particular molecular interactions to the TCF. For example, in the mechanistic studies of the dopamine transporter or TMEM16 family lipid scramblases, we demonstrated that TCFs could be constructed in the context of microstate free energies inferred by the Markov State Model analysis. For more details, please refer to the following publication: Khelashvili G, Kots E, Cheng X, Levine MV, Weinstein H. The allosteric mechanism leading to an open-groove lipid conductive state of the TMEM16F scramblase. Communications Biology 2022 5(1):990 PMID 36123525

Continuum-Molecular Dynamics (CTMD) software

The CTMDapp software calculates the deformation profiles of the bilayer and the free energy cost of the membrane deformation around multi-segment transmembrane proteins, taking into account the radially non-uniform hydrophobic surface of the protein. For more details, please refer to the following publication: Mondal S, Khelashvili G, Shan J, Andersen OS, Weinstein H. Quantitative modeling of membrane deformations by multihelical membrane proteins: application to G-protein coupled receptors. Biophysical Journal 2011 101:2092-2101 PMID 22067146

Real-space fluctuations (RSF) method to quatify material properties of lipid membranes

The RFS method is a computational approach for quantification of the elastic properties of lipidic assemblies of arbitrary shape and composition (including lipid mixtures) from the analysis of molecular dynamics simulations. The method is applicable to diverse lipid assembly types, ranging from bilayers in the liquid ordered and disordered phases to a study of the inverted hexagonal phase. For more details, please refer to the following publication: Doktorova M, Harries D, Khelashvili G. Determination of bending rigidity and tilt modulus of lipid membranes from real-space fluctuation analysis of molecular dynamics simulations. Physical Chemistry Chemical Physics 2017, 19:16806-16818 PMID 28627570

Self-assembled lipid cubic phases

Structures of cubic phases of various types of monoolein lipids self-assembled during molecular dynamics simulations performed under different hydration conditions can be downloaded, together with the corresponding Martini force-field parameter files used for these simulations. For more details, please refer to the following publication: Johner N, Mondal S, Morra G, Caffrey M, Weinstein H, Khelashvil G. Protein and Lipid Interactions Driving Molecular Mechanisms of in meso Crystallization. Journal of the American Chemical Society 2014 136(8):3271-84 PMID 24494670

  • George Khelashvili, Ph.D.

    Associate Professor at the Department of Physiology and Biophysics, Associate Professor in Computational Biophysics at Weill Cornell Medicine in New York.

Contact