A novel strategy for classifying the output from an in silicovaccine discovery pipeline for eukaryotic pathogens using machine learning algorithms

Table 2 High-throughput standalone programs used in this study to predict protein characteristics

Name	Version	Predicted protein characteristic	URL (last viewed November 2013)	Published accuracy^a
WoLF PSORT	0.2	Protein localisation	http://wolfpsort.org/WoLFPSORT_package/version0.2/	80.0% [11]
SignalP	4.0	Secretory signal peptides	http://www.cbs.dtu.dk/services/SignalP/	93.0%^b[9]
TargetP	1.1	Secretory signal peptides	http://www.cbs.dtu.dk/services/TargetP/	90.0% [10]
TMHMM	2.0	Transmembrane domains	http://www.cbs.dtu.dk/services/TMHMM/	97.0% [13]
Phobius	_	Transmembrane domains and signal peptides	http://phobius.binf.ku.dk/instructions.html	94.1% [12]
Peptide-MHC I Binding^c		Peptide binding to MHC class I	http://tools.immuneepitope.org/main/html/download.html	95.7%^d[14]
Peptide-MHC II Binding^c		Peptide binding to MHC class II	http://tools.immuneepitope.org/main/html/download.html	76.0%^d[15]

^aPredictive accuracies taken from publications by the creators of the programs. The prediction accuracy varies for different target pathogens.
^bSignalP version 3.0.
^cPrediction Tools from The Immune Epitope Database and Analysis Resource (IEDB) [http://www.iedb.org].
^dArea under curve value (AUC). Program uses different methods. For MHC I best method = artificial neural network (ANN) [14] and MHC II best method = Consensus [15].

ISSN: 1471-2105