Lu Zhu (started October 2014)
Meaningful data and methods to reliably and systematically study protein subcellular localization (SCL) and exploration of network of protein-protein interactions (PPI) have long been the goals of functional biology and the post genomic era for understanding the mechanistic hypotheses of a cell. Each type of cellular compartment provides a unique physiological environment within which specialized functions are carried out.To interact, proteins (or any other molecules) must necessarily share a common SCL, or an interface between physically adjacent SCLs, transiently or conditionally. Several PPI based SCL prediction methods have been developed using different approaches and data sources. When predicting the SCL of a given protein, not only the protein itself, but also all its interacting partners should be considered. Predicting multiple localizations is always a challenge. Some of the existing method made some process. All of the existing methods are designed to predict all the localizations the same time and use the same parameters for maximizing the global prediction performance. This leads to the poor prediction accuracy for certain SCLs. In our method apply MRF analysis to PPI network for protein SCL assignment problem. Each protein SCL is individually predicted using a MRF model with suitable parameter configuration corresponding to this SCL. We construct the MRF models, to predict if an unknown protein locates in a particular SCL, we consider (1) the prior probability of any protein locates in this SCL, (2) the number of interacting neighbors located in the same SCL, (3) the number of interacting neighbors located in the adjacent SCLs, and (4) the sequence-based features of the protein. A value of the probability of a protein locates in a certain SCL is assigned for every protein and every SCL.
Supervisors: Ralf Hofestädt (Bielefeld University), Martin Ester (Simon Fraser University)