Subcellular localization prediction for human internal and organelle membrane proteins with projected gene ontology scores.

The membrane proteins make up more than a third of all known human proteins. The subcellular localizations play a key role to elucidate the potential biological functions of these membrane proteins. Although the experimental approaches for determining protein subcellular localizations exist, they are usually costly and time consuming. Thus, computational predictions provided an alternative approach for determining the protein subcellular localizations. However, current subcellular location predictors are generally developed for globular proteins. They did not perform well for membrane proteins. In this paper, we proposed a novel prediction algorithm, namely Projected Gene Ontology Score, which introduces the Gene Ontology annotation as a descriptor of the protein. This algorithm could significantly improve the prediction accuracy for the subcellular localizations of membrane proteins. It can designate each protein to one of the eight different locations, while the existing algorithm only covers three locations. Actually, the biological problem considered by our algorithm goes one level deeper than the existing algorithms. In addition, our algorithm can provide more than one location for the testing protein, which could be very useful in practical studies. Our algorithm is expected to be a good complement to the existing algorithms and has the potential to be extended to solve other problems.