Integrated Bioinformatic Approaches to the Prediction of Protein-Protein Interactions

Abstract

Protein-protein interactions (PPIs) play a fundamental role in many biologicalprocesses such as signal transduction from the extracelluar space to cytosol. Functionsof less characterized proteins can often be deduced from PPI networks. Varioussequence-based approaches were taken to predicting and understanding potential PPIsusing bioinformatic means. Initially, the mirrortree method was comprehensivelyexamined to derive a robust approach for PPI predictions. The analysis has revealed thatmirrortree is extremely sensitive to many factors especially sequence diversity and theselection of orthologues. Indeed, higher sequence diversity improves the predictivepower of the approach. In an attempt to improve prediction accuracy, various speciationsignal correction methods were evaluated and the RNA-based approaches appear to bemore effective in removing the speciation signal and ultimately produce more accuratepredictions. The utility of mirrortree was further extended for domain-domaininteractions in fibrillin-1. However, due to the low sequence diversity of theorthologues, poor prediction results were obtained. Furthermore, a residue basedmethod utilizing the mutual information (MI) statistic was evaluated for intramolecularprotein interaction predictions. Similar to the mirrortree method, removal of thebackground signal occurring from common ancestry improves the prediction accuracy.When MI of a third position was incorporated to facilitate the interaction predictionbetween two contacting positions, the prediction quality was increased. Moreover, inorder to identify clusters consisting of three contacting residues, position combinationswith the highest significant partial correlation coefficients were extracted and theiratomic distances were compared to assess the accuracy of the prediction. Lastly, ananalysis was carried out to study the association between PRINTS fingerprints andfunctionally important interaction sites in seven G protein-coupled receptor families.More than 50% of the functional sites acquired from literature were found to be in closeproximity to fingerprint motifs. In the surface patch analysis, over 80% of the functionalsites were shown to overlap a motif cluster. Overall, the approaches taken in this thesishave tackled interaction predictions from various directions and keenly provide someinsights for protein-protein interactions and evolution.

Date of Award	31 Dec 2011
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Simon Hubbard (Supervisor) & Terri Attwood (Supervisor)