In silico prioritisation of candidate genes for prokaryotic gene function discovery: an application of phylogenetic profiles

Table 2 CGP performance on peptidoglycan-related genes (Escherichia coli K-12, 4131 genes).

	Validation sets
Methods	C (8 genes)		B (28 genes)		M (51 genes)
	AUC	(/η _max)	AUC	(/η _max)	AUC	(/η _max)
Statistical CGP (scoring functions)
sens	0.913	(2.5/10.6)	0.891	(2.3/6.0)	0.818	(1.9/4.2)
spec	0.321	(0.4/1.4)	0.310	(0.4/1.2)	0.418	(0.8/2.0)
ppv	0.405	(0.8/5.2)	0.423	(1.2/18.4)	0.553	(1.7/28.6)
npv	0.974	(3.9/42.0)	0.956	(3.5/20.9)	0.891	(2.8/13.2)
amss	0.989	(4.8/110.)	0.966	(4.1/53.7)	0.911	(3.5/44.7)
hmss	0.989	(4.9/113.)	0.969	(4.2/55.3)	0.909	(3.5/45.6)
OR	0.403	(0.8/5.2)	0.424	(1.2/18.4)	0.552	(1.7/28.6)
chisq	0.984	(4.7/73.8)	0.963	(3.9/35.9)	0.902	(3.2/27.0)
bchisq	0.984	(4.7/73.8)	0.963	(3.9/35.9)	0.903	(3.2/27.0)
F	0.965	(4.0/45.8)	0.921	(3.2/22.5)	0.838	(2.5/15.1)
Inductive CGP (machine learning algorithms)
NB	0.930		0.889		0.820
LR	0.882		0.935		0.828
ADTree	0.976		0.981		0.925
IBk	0.998		0.929		0.946
J48	0.935		0.828		0.752
SMO/Poly	0.997		0.876		0.933
SMO/RBF	0.963		0.932		0.964

This table lists the performance of statistical and inductive CGP in prioritising peptidoglycan-related genes in Escherichia coli K-12. Abbreviations: sens: sensitivity; spec: specificity; ppv: positive predictive value; npv: negative predictive value; amss: arithmetic mean of sensitivity and specificity; hmss: harmonic mean of sensitivity and specificity; OR: odds ratio; chisq: chi-square; bchisq: signed chi-square; F: F-measure; NB: naïve Bayes classifier; LR: logistic regression; ADTree: alternating decision tree; IBk: k-nearest neighbour classifier; J48: J48 decision tree; SMO: support vector machine trained by sequential minimal optimisation algorithm; Poly: polynomial kernel; RBF: radial basis function kernel.

ISSN: 1471-2105