Performance analysis of protein structure clustering techniques and CUDA implementation of RMSD computation

Knowledge of the 3-dimensional structure of proteins is an important aspect in the field of structure based drug design. Structure prediction algorithms generally operate by computationally generating a large number of protein structures known as decoys and selecting the best candidates from among them. This is done by clustering the decoy set to identify the best models. RMSD (Root Mean Square Deviation) is the metric used for measuring similarity between protein structures. As the number of decoys becomes larger, the huge computational time of RMSD calculation affects the overall performance. This paper is about the performance analysis done on clustering techniques of SPICKER, Calibur and Hierarchical Ward's clustering and also describes 2 methods for parallelizing RMSD computation.