Biomedical Sciences and Engineering MS Thesis Defense by Efe Elbeyli







Title: Assessment of PRISM with previous CAPRI rounds


Speaker: Efe Elbeyli


Time: November 16, 2017, 13.00


Place: ENG 208

Koç University

Rumeli Feneri Yolu

Sariyer, Istanbul

Thesis Committee Members:

Prof. Özlem Keskin (Advisor, Koc University)

Prof. Attila Gürsoy (Co-Advisor, Koc University)

Prof. Burak Erman (Koc University)

Prof. İ. Halil Kavaklı (Koc University)

Assist. Prof. S. Ece Özbabacan (İstanbul Medeniyet University)



Proteins are key elements of a cell to perform the wide range of molecular and cellular activity. Proteins perform their function through binding to other proteins, DNA, RNA, and small molecules. Therefore, predicting how a protein interacts with its binding partners is one of the most important objectives of structural biology. As a result of the improvements in experimental structure determination methods, the number of individual protein structures in PDB has increased vastly. However, the number of complex structures does not increase as fast as the individual proteins which creates demand for new approaches to predict complexes from the individual protein structures. Currently, there are a lot of computational approaches to predict the complex structures. Critical Assessment of Prediction of Interactions (CAPRI) is a well-known community-wide experiment with the purpose of establishing a routine which allows testing the performance of several different docking algorithms created. The success rate of predicted structures is verified by the several evaluation criteria determined by the CAPRI association. Those criteria are interface and ligand rmsds (I-rmsd, L-rmsd), native residue contacts and number of clashes. In this study, I assessed the performance of PRISM (Protein Interaction by Structural Matching) using CAPRI evaluation criteria. The main objectives were to determine how much PRISM is successful in predicting the complex structures of available CAPRI targets and to compare PRISM with other available docking algorithms. PRISM could not predict the correct complex structures for 33% of targets majority of which correspond to homodimers and enzyme/inhibitor complexes. The results also indicate that considering just the structures with the negative energy score results in the loss of 36% of successful predictions implying a problem in the scoring function. For further testing, RosettaDock was used as an alternative scoring function. Both scoring approaches yielded a correlation. To increase the success rate, some parameters of PRISM were changed however, no significant improvement has been achieved.