Structure-based prediction of protein-protein interaction sites

摘要: Protein-protein interactions play a central role in the formation of protein complexes and biological pathways that orchestrate virtually all cellular processes. Reliable identification specific amino acid residues form interface with one or more other proteins is critical to understanding structural physico-chemical basis their key processes, predicting complexes, validating predicted by high throughput methods, identifying prioritizing drug targets computational design. Because difficulty cost experimental characterization residues, there an urgent need for methods reliable protein-protein from sequence, when available, structure query protein, known, its putative interacting partner. Against this background, thesis develops improved interfaces three dimensional unbound without considering information binding Towards end, we develop (i) ProtInDb (http://protindb.cs.iastate.edu), database facilitate (a) generation datasets can be used perform analysis interaction sites train evaluate predictors (b) visualization between both sequences 3D structures, among applications; (ii) PoInterS (http://pointers.cs.iastate.edu/), method formed spatially contiguous clusters based on predictions generated residue predictor. divides surface into series patches composed several uses outputs rank select small set are most likely constitute sites; (iii) PrISE (http://prise.cs.iastate.edu/), similarity element neighboring elements extracted non-interface regions members experimentally determined complexes. A captures atomic composition solvent accessibility closest neighbors structure. decomposes searches similar large belong The each then infer whether not residue. results our experiments using variety benchmark show generally outperform state-of-the-art structure-based respectively.