Protein-protein etkileşimlerinin otomatik tahmini için yüksek başarımlı bir algoritma

ABSTRACT The major goal of Computational Biology and Bioinformatics is to achieve a better un derstanding of the principles of biological systems and processes using informatics tools. Elucidation of the full network of protein-protein interactions is a crucial part of this chal lenge. Thus, there is a growing need for fast and reliable in silico methods for predicting protein-protein interactions. Here, we present a high-performance algorithm for automated prediction of protein-protein interactions. We adopt a novel bottom-up approach that com bines structure and sequence conservation in protein interfaces. Starting with 67 known structures of protein interfaces and 6170 protein structures, we predicted 62616 distinct in teractions. We then checked whether these interactions existed in three different interaction databases. We also searched literature for some interesting cases. The results displayed a good balance of verified and unverified predictions. Verified interactions prove the relia bility of our algorithm whereas unverified ones may correspond to unobserved interactions that actually occur in nature or may synthetically be realized in laboratory conditions. We believe these unverified predictions may have important implications regarding drug design. We parallelized the algorithms to reduce execution times from the order of months to days: parallelized prediction algorithm demonstrated a speed up of 29.39 on a 32 node Beowulf cluster. IV