Prediction of signal peptide in proteins with neural networks

Abstract

In this paper we present a new Neural-Network-based predictor trained and tested on a set of well annotated proteins to tackle the problem of predicting the signal peptide in protein sequences. The method trained on a set of experimentally derived signal peptides from Eukaryotes and Prokaryotes, identifies the presence of the sorting signal and predicts their cleavage sites. The accuracy in cross-validation is comparable with previously presented programs reaching the 97%, 96% and 95% for Gram negative, Gram positive and Eukaryotes, respectively. © Springer-Verlag Berlin Heidelberg 2003.