Protein etkileşimlerinin geniş ölçekli nitelendirilmesi: Protein-protein arayüzlerindeki sıcak noktaların ve uzaysal desenlerin belirlenmesi

Complex biological processes in cells are carried out by association of biological molecules including protein-protein interactions. Many diverse approaches have been developed to explain association of proteins. Notwithstanding, mechanisms of protein-protein interactions are still not adequately elucidated. In order to characterize protein-protein interfaces --the architectural binding site elements in between two monomers-- on a large scale, here, we present novel computational techniques addressing two important structural biology problems: prediction of interface hot spots and discovery of spatial interface motifs.Hot spots are residues comprising only a small fraction of interfaces yet accounting for the majority of the binding energy. We present a new efficient method to determine computational hot spots on protein interfaces based on sequence conservation and solvent accessibility of interface residues. The predicted hot spots are observed to correlate considerably with the experimental hot spots. The results reveal, due to lack of available experimental data, machine learning approaches do not overperform proposed empirical approach. Predicted computational hot spots on protein interfaces can be queried and visualized via HotSprint web interface located at http://prism.ccbb.ku.edu.tr/hotsprint .Protein associations might be structurally mediated by interacting patterns on the protein binding sites (spatial interface motifs). In the second part of this thesis study, a new frequently reoccurring interface spatial pattern discovery method employing graph mining is developed. Proposed method, sPpprint, finds not necessarily sequence-contiguous and a priori unknown common set of atoms (substructures) on the protein interfaces. Initial results suggest that there exist discriminative spatial protein interface motifs that may be used to determine type of the interface.