Structural complex prediction based on protein interface recognition

Esmaielbeiki, Reyhaneh (2013) Structural complex prediction based on protein interface recognition. (PhD thesis), Kingston University, .


This dissertation contributes to the state of the art in protein interface prediction and detection of native-like docked poses by re-ranking them using protein interface knowledge. We started by investigating binding site patterns among homologues of a target protein in order to create a 3D motif. This structural binding site descriptor enables the re-ranking of docked complexes of the target protein. Although 3D motifs provide biological insight of protein interactions and have usage in real applications, they are not suitable for high through-put analysis. Therefore, we introduced a novel protein interface prediction framework which uses a weighted scoring schema to detect interface residues of the target protein using its homologues. The weights quantify both homology closeness between the target protein and its homologues and the diversity between the interacting partners of these homologues. The main novelty of this predictor is that it takes into account the nature of homologues interacting partners. It was further exploited for the development of a method for re-ranking docked conformations using predicted interface residues. We have evaluated both our interface predictor and re-ranking of docked poses using standard benchmarks. Comparisons to current state-of-the-art methods reveal that the proposed approaches outperform all their competitors. However, similarly to current interface predictors, our framework does not explicitly refer to pairwise residue interactions which leaves ambiguity when assessing quality of complex conformations. In addition, the performance of our interface predictor generally does not outperform the best available homologue interfaces if it was used as prediction. Therefore, we investigated the detection of the best homologue using the 'binding site transitivity' concept: given two query protein chains, interfaces of the first query protein are structurally compared against binding sites of the homologues' partners of the second query chain. This method not only allows detection of the best homologue for a reasonable number of proteins but also produces a docked structure of the two query chains. Finally, experiment suggests a meta interface predictor combining the prediction of our former interface predictor with the latter predictor based on binding site transitivity could further improve interface prediction.

Actions (Repository Editors)

Item Control Page Item Control Page