Clustering of Protein Conformations Using Parallelized Dimensionality Reduction

—Ascertaining the conformational landscape of a macromolecule, like protein is indispensable to understanding its characteristics and functions. In this work, an amassment of these techniques is presented, that would be an aid in sampling of these conformations better and faster. The datasets that represent these conformational dynamics of proteins are complex and high dimensional. Therefore, there arises a need for dimensionality reduction methods that best conserve the variance and further the analysis of the data. We present a parallelized version of a well-known dimensionality reduction method, Isomap. Isomap has been shown to produce better results than linear dimensionality reduction in approximating the complex landscape of protein folding. However, the algorithm is compute-intensive for large proteins or a large number of samples, used to model a path that a protein undergoes. We present an algorithm, parallelized using OpenMP, with a speed-up of approximately twice. The results are in agreement with the ones obtained using sequential Isomap.