Mixture models for protein structure ensembles

摘要: Motivation: Protein structure ensembles provide important insight into the dynamics and function of a protein contain information that is not captured with single static structure. However, it clear priori to what extent variability within an ensemble caused by internal structural changes. Additional results from overall translations rotations molecule. And most experimental data do relate structures common reference frame. To report meaningful values intrinsic dynamics, precision, conformational entropy, etc., therefore disentangle local global heterogeneity. Results: We consider task disentangling heterogeneity as inference problem. use probabilistic methods infer missing on frames stable sub-states. this end, we model mixture Gaussian probability distributions either entire conformations or segments. learn these models using expectation–maximization algorithm. Our first can be used find multiple conformers in ensemble. The second partitions chain locally segments core elements less structured regions typically found loops. Both are simple implement only free parameter: number analyse ensembles, molecular trajectories change proteins. Availability: Python source code for analysis available authors upon request. Contact: michael.habeck@tuebingen.mpg.de