Molecular simulations of proteins give rise to a large amount of data which makes identifying important details by mere visual inspection difficult. By using a wide range of data-driven tools such as neural networks, clustering and Gaussian mixture density estimation, we are not only able to identify important details in the simulations but we also reduce the risk of introducing bias to the analysis.
We develop tools for estimating free energy landscapes, as well as clustering well-defined core-states. We also adapt and apply machine learning methods for characterizing intermediate states or differences between conditions, e.g. ligand-bound vs ligand-unbound. In addition to this, we use graph theory and network analysis to understand allosteric communication in membrane proteins.
Our tools and tutorials can be found at
Molecular Insights From Conformational Ensembles Using Machine Learning, O. Fleetwood, M.A. Kasimova, A.M. Westerlund, L. Delemotte, Biophys. J., 118 (3), pp 765-780
InfleCS: Clustering Free Energy Landscapes with Gaussian Mixtures, A.M. Westerlund and L. Delemotte, J. Chem. Theory Comput., 15 (12) 6052-6059,
Effect of Ca2+ on the promiscuous target-protein binding mechanism of calmodulin A.M. Westerlund and L. Delemotte, PloS Comp Biol, DOI: 10.1371/journal.pcbi.1006072 2018
Inference of Calmodulin’s Ca2+-Dependent Free Energy Landscapes via Gaussian Mixture Model Validation A.M. Westerlund, T.J. Harpole, C. Blau and L. Delemotte, J. Chem. Theory Comput. DOI: 10.1021/acs.jctc.7b00346 2018
Back to Membrane-Protein Modeling