Classification and Scoring of Protein Complexes

Abstract

Proteins interactions mediate all biological systems in a cell; understanding their interactions means understanding the processes responsible for human life. Their structure can be obtained experimentally, but such processes frequently fail at determining structures of protein complexes. To address the issue, computational methods have been developed that attempt to predict the structure of a protein complex, using information of its constituents. These methods, known as docking, generate thousands of possible poses for each complex, and require effective and reliable ways to quickly discriminate the correct pose among the set of incorrect ones. In this thesis, a new scoring function was developed that uses machine learning techniques and features extracted from the structure of the interacting proteins, to correctly classify and rank the putative poses. The developed function has shown to be competitive with current state-of-the-art solutions.