Prediction of protein structures and protein-protein interactions: a bioinformatics approach

Prediction of protein structures from sequences and protein-protein interaction from structures are two grand challenges for computational biologists in the genomic era. Both remain unsolved problems despite considerable effort. Thus, simplification of the problems, i.e. prediction of solvent accessibility (exposed or buried) and protein-protein interface residues can be useful as a first step and aid the structure modeling processes.We refined of a suite of five methods in solvent accessibility prediction over a large dataset and proposed a metamethod based on an ensemble average of the individual methods, leading to a two-state classification accuracy of 80%. Our results are 0.7-6.7% better than current state-of-art methods. Results suggest solvent accessibility prediction could be a valuable tool in improving protein structure prediction. We applied these methods for predicting sites of deleterious mutations and 80% accuracy was achieved, suggesting that the methods for prediction of solvent accessibility can be the basis for accurate predictions of deleterious mutations.We developed of a robust program, cons-PPISP, for predicting interface residues in protein-protein complexes. Having successfully improved the prediction accuracy from 70% to 80%, we showed potentially it can complement experimental methods such as chemical-shift perturbations in characterizing protein-protein interfaces. Incorporation of cons-PPISP predictions in a docking program in the latest CAPRI rounds demonstrated its ability in guiding the docking process.Ph.D., Computational Biophysics -- Drexel University, 200