Computational prediction of protein subcellular localization and function

Official URL: http://risc01.sabanciuniv.edu/record=b1064374 (Table of Contents)

In this study, we present a computational approach in which it is possible to directly predict the protein functional categories from sequence and to identify the protein subcellular localization, which, in turn, is helpful for functional classification. Subcellular protein locations and functions have been predicted basically from amino acid composition by using a machine learning approach. Expert systems based on Support Vector Machines have been designed to predict subcellular locations for proteins both in plants and nonplants, and function particularly for nonplants. Four subcellular localization categories for plant and nonplant proteins have beenidentified by correct prediction accuracies of 95.4%, and 99.7% respectively. In addition to the three common categories mitochondrial, extracellular / secretory, and nuclear; the classes cytosolic for nonplants, and, chloroplast for plants are included. Functional categories related to the subcellular compartments are predicted by using a similar approach applied for localization prediction. 92.9% of the 2321 protein sequences have been correctly assigned into the selected 10 functional categories. Finally, the contribution of the data-mining of the MEDLINE papers to the function prediction is tested by another protein data set.