Improvement of PSORT II Protein Sorting Prediction for Mammalian Proteins

摘要

The PSORT system is a unique tool for the prediction of protein subcellular localization in a sense that it can deal with proteins localized at almost all the subcellular compartments. In its several versions, PSORT II was developed for the predictionof eukaryotic proteins using yeast sequences as its training data. The reason why the data from a single species were used was that training data were favored to reflect the subcellular proportion of a proteome. However, since the yeast is a unicellularorganism, applying PSORT II to sequences of multicellular organisms can be problematic sometimes. For example, it has been pointed out that secretary proteins tend to be under-predicted. Since the first release of PSORT II, genome projects have been producing rich information of genes for many model organisms including yeasts, nematode, mouse and human. Amongst them, the information of mouse genes is managed in Mouse Genome Database (MGD) and Mouse Genome Informatics (MGI). The information includes thedata of the full length cDNAs with the annotation of subcellular localization sites of their products. In this work, we report the improvement of PSORT II from three aspects: the employment of mammalian (murine) data, the optimization of the learning method, and the optimization of the sequence features used.