SOSUImp1: high performance prediction system for single-spanning membrane proteins

Single-spanning membrane proteins (MP1) occupy the largest component of membrane proteins in total open reading frames of organisms, having essential functions such as signal transduction, immunological reaction and cell adhesion. We developed a novel software system comprised of two filtering layers for predicting MP1 with or without a signal peptide region. In the first filtering layer, we selected membrane proteins with one or two transmembrane (TM) regions by the membrane protein prediction system SOSUI, which is accurate in predicting transmembrane regions but cannot identify signal peptide regions. The second filtering layer was comprised of several modules for distinguishing signal peptide regions. On the assumption that a signal peptide has two kinds of sequences at the N-terminus by which the signal peptide is embedded into membrane and cleaved at its C-terminal end, we calculated two discrimination scores by the canonical discriminant analysis, using averages of several physical properties around the first N-terminal hydrophobic cluster. This prediction system SOSUImp1 comprised of two filtering layers could discriminate very accurately among five types of proteins: cytoplasmic soluble proteins and secretory proteins, MP1 with and without a signal peptide, and multi spanning membrane proteins. The performance for MP1 with a signal peptide that is important in the cell-cell communication was particularly high compared with previous prediction systems.The prediction system SOSUImp1 and the dataset of 5932 proteins used for developing the system are available at http://bp.nuap.nagoya-u.ac.jp/sosui/mp1/