| Hi-index | 3.84 |
Motivation: Subcellular localization is a key feature in the process of functional annotation of both globular and membrane proteins. In the absence of experimental data, protein localization is inferred on the basis of annotation transfer upon sequence similarity search. However, predictive tools are necessary when the localization of homologs is not known. This is so particularly for membrane proteins. Furthermore, most of the available predictors of subcellular localization are specifically trained on globular proteins and poorly perform on membrane proteins. Results: Here we develop MemLoci, a new support vector machine-based tool that discriminates three membrane protein localizations: plasma, internal and organelle membrane. When tested on an independent set, MemLoci outperforms existing methods, reaching an overall accuracy of 70% on predicting the location in the three membrane types, with a generalized correlation coefficient as high as 0.50. Availability: The MemLoci server is freely available on the web at: http://mu2py.biocomp.unibo.it/memloci. Datasets described in the article can be downloaded at the same site. Contact: casadio@biocomp.unibo.it Supplementary information:Supplementary data are available at Bioinformatics online.