Improvements of protein function annotations using structures

Understanding the function of each gene in the genome is central to life sciences. A large number of high resolution protein structures provide new opportunities for computational and structural biologists. My thesis work is dedicated to elucidating protein functions from protein structures using computational methods. The first part of my work focused on improving sequence based functional site prediction by combining sequence conservation and ROSETTA computational protein design. This method has achieved improved functional site identification accuracy when compared with related work. The web service based on this work is widely used by top research institutes globally. The second part of my research focus aimed to elucidate protein functions using structures as the only source of information. Specifically, I first developed a geometrical knowledge-based scoring function to model protein/ion interactions accurately. Combined with a novel high performance grid based sampling method, protein/ion binding sites can be predicted at high resolution within minutes. Subsequently, I extended the scoring function to handle ion mediated protein/substrate interactions. Then I assembled a carbohydrate rotamer library to annotate lectin functions from solved lectin structures. All of this work demonstrats that protein structures are extremely useful for protein function annotations and more efforts should be put into this field to continue the research described above.