Characterization of Molecular Dynamic Trajectory Using K-means Clustering

Abstract

Conformations of kinase obtained from molecular dynamic (MD) simulation plays an important role in molecular docking experiment in the field of drug discovery and development. The scanning of all the MD conformations against millions of drug like molecules is not feasible as it requires high computational cost. Clustering techniques have been widely explored to reduce the conformations into manageable size. Clustering is the Artificial Intelligence based technique in which conformations are partitioned into clusters that exhibit similar behavior thus reducing the scanning cost. This paper analyzed two techniques Principal component analysis and K-means clustering to analyze the molecular simulation data. The clusters thus obtained are evaluated by using Average Silhouette Width and Calinski-Harabasz Index by considering varying number of clusters from 2 to 10 by varying Principal Components from 4 to 6. The result obtained by the purposed framework shows that Average Silhouette Width for K-means clustering is independent of Principal Component subspace.