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

Table 1 Datasets used for training and testing machine learning models

Name^a	Number of proteins in each group^b			Organism	Comments
	Membrane-associated	Secreted	Neither membrane-associated nor secreted
T. gondii	8	13	18	Toxoplasma gondii
Plasmodium	47	26	51	Plasmodium	Includes falciparum, yoelii yoelii, and berghei
C. elegans	324	56	380	Caenorhabditis elegans
Combined species	379	95	449	Combination of organisms	Includes T. gondii, C. elegans, P. falciparum, P. yoelii yoelii, and P. berghei
Benchmark	70^c		70	Combination of two organisms	T. gondii and Neospora caninum (excludes the proteins in T. gondii dataset)

^aThis is the name used to refer to the dataset throughout the paper.
^bProteins (except for the benchmark dataset) were initially grouped in accordance with the subcellular location descriptor in UniProtKB, then fine-tuned in accordance to cross-validation testing, epitope presence, and reference to other UniProtKB annotations and Gene Ontology. Benchmark proteins were taken from published studies (70 experimentally shown to induce immune responses).
^cCombination of proteins from membrane-associated, secreted, and unknown subcellular locations.
Note: Membrane-associated and Secreted proteins are expected ‘YES’ classification for vaccine candidacy. Neither membrane-associated nor secreted proteins are expected ‘NO’ classification. There was an attempt to create an equal representation of YES and NO classifications in the training datasets.

ISSN: 1471-2105