Journal of Proteomics & Bioinformatics

Journal of Proteomics & Bioinformatics
Open Access

ISSN: 0974-276X

+44 1223 790975


Evaluation of Computational Methods for Secreted Protein Prediction in Different Eukaryotes

Xiang Jia Min

Secreted proteins play important biological roles in eukaryotes. Computational identification of all secreted proteins, i. e. the secretome, from predicted proteome of completely sequenced genomes is an essential step in functional annotation. To develop screening methods for secreted proteins in different kingdoms of eukaryotes, we have evaluated the prediction accuracies of SignalP, Phobius, TargetP, and WolfPsort used individually or in combination with TMHMM and PS-Scan. Prediction accuracy was represented by Mathews’ Correlation Coefficient (MCC). The tools show different strength for predicting secreted proteins in different kingdoms of eukaryotes. When individual tools were used, we found that the tools having the highest accuracy were WolfPsort for fungi (73.1%), Phobius for animals (82.8%), SignalP for plants (55.4%), and Phobius for protists (42.1%), respectively. Except using Phobius, combining the prediction tools with TMHMM significantly improved the prediction accuracy in all data sets. Based on the measured accuracies, it is recommended that using the following methods for secretome prediction in different eukaryotes: SignalP/TMHMM/ WolfPsort/Phobius/PS-Scan for fungi (83.4%), Phobius/WolfPsort/PS-Scan for animals (86.7%), SignalP/TMHMM/ Phobius/TargetP/PS-Scan for plants (73.2%), and combining all the tools for protists (52.8%).