CELLO: A subCELlular LOcalization predictor

CELLO is a multi-class SVM classification system. CELLO uses 4 types of sequence coding schemes: the amino acid composition, the di-peptide composition, the partitioned amino acid composition and the sequence composition based on the physico-chemical properties of amino acids. We combine votes from these classifiers and use the jury votes to determine the final assignment. The general architecture of our predictive system is shown below.

Comparison of CELLO with other methods

Table 1 The comparison of predictive performances of different approaches in the prediction of subcellular localization for eukaryotic sequences (accuracy is in %)

	CELLO	Reinhardt & Hubbard	Yuan	SubLoc
Cytoplasmic	85.1	55	78.1	76.9
Extracellular	84.3	75	62.2	80.0
Mitochondrial	63.2	61	69.2	56.7
Nuclear	96.0	72	74.1	87.4
Accuracy	87.0	66	73.0	79.4

Table 2 The comparison of predictive performances of different approaches in the prediction of subcellular localization for Gram-negative bacteria (accuracy is in %)

Localization	CELLO		PSORT-B		PSORT I		SubLoc
Localization	Accuracy	MCC	Accuracy	MCC	Accuracy	MCC		Accuracy	MCC
Cytoplasmic	90.7	0.85	69.4	0.79	75.4	0.58		75.0	0.74
Inner membrane	88.4	0.92	78.7	0.85	95.1	0.64		82.8	0.89
Periplasmic	86.9	0.80	57.6	0.69	66.4	0.55		68.9	0.71
Outer membrane	94.6	0.90	90.3	0.93	54.5	0.47		89.1	0.86
Extracellular	78.9	0.82	70.0	0.79	–	–		69.5	0.78
Accuracy	88.9	–	74.8	–	60.9	–		78.5	–

References

1) Yu CS, Lin CJ, Hwang JK: Predicting subcellular localization of proteins for Gram-negative bacteria by support vector machines based on n-peptide compositions. Protein Science 2004, 13:1402-1406.

2) Yu CS, Chen YC, Lu CH, Hwang JK: Prediction of protein subcellular localization. Proteins: Structure, Function and Bioinformatics 2006, (in press).

3) Yu CS, Lin CJ, Hwang JK: Prediction of Subcellular Locations by Support Vector Machines Using Multiple Feature Vectors Based on n-peptide Compositions. (unpublished data).