Clustering algorithms applied on analysis of protein molecular dynamics

Analysis of molecular dynamic (MD) simulation has been difficult since this method generates a lot of conformations. Thus clustering algorithms have been applied to group similar structures from MD simulations, but the choice of the information to be clustered is still a challenge. In this work, we propose the use of Euclidean distance matrices (EDM) from conformations as input data to clustering algorithms. We used approaches combining non-reduction or reduction of data dimensionality (MDS and isomap methods), and different clustering algorithms (k-means, ward, mean-shift and affinity propagation). Results indicated that EDM could be a good information to be used in clustering conformations from MD. For data with small protein structure variation, the mean-shift algorithm had good results in both non-reduced and reduced data. However, for data with large protein structure variation, the methods that work better with smooth-density data (k-means and ward) had good results.