MD Simulations Analysis

 

Molecular dynamics trajectories of proteins give rise to a large amount of data which makes identifying properties by mere visual inspection difficult. Unsupervised learning and data analysis allows to automatically extract important features from simulations. We apply protocols using spectral clustering with data analysis to identify and characterize intermediate states of a conformational ensemble [1], we develop accurate free energy landscape estimation methods based on Gaussian mixture models [2], we graph theoretical concepts to describe the molecular mechanisms underpinning protein function. We also resort to neural networks trained to distinguish between different conformational states and apply layer-wise relevance propagation based on deep Taylor decomposition to identify the most important residues for classification.

Recent publications:

[1] Effect of Ca2+ on the promiscuous target-protein binding mechanism of calmodulin AM Westerlund, L Delemotte, PloS Comp Biol, DOI: 10.1371/journal.pcbi.1006072

[2] Inference of Calmodulin’s Ca2+-Dependent Free Energy Landscapes via Gaussian Mixture Model Validation AM Westerlund, TJ Harpole, C Blau and L Delemotte, J. Chem. Theor. Comp. DOI: 10.1021/acs.jctc.7b00346

Contributors:

Back to Membrane-Protein Modeling