The use of genetic programming in the analysis of quantitative gene expression profiles for identification of nodal status in bladder cancer

Table 4 Probability of gene usage from the set of 70 genes due to random chance.

Gene	Actual occurrence of gene (in 220 rules)	Binomial probability	p-value (one-sided)
KDR	159	9.69E-130	<0.00001
MAP2K6	146	1.13E-110	<0.00001
ICAM1	121	4.10E-78	<0.00001
ANXA5	60	7.04E-20	<0.00001
CDK8	56	3.38E-17	<0.00001
CDKN1A	55	1.49E-16	<0.00001
JUNB	49	6.56E-13	<0.00001
TGFBR2	41	1.08E-08	<0.00001
TNF	24	1.11E-02	0.01490
TNFAIP1	23	1.76E-02	0.02599
CCND3	19	6.73E-02	0.16020
PDGFRL	18	8.23E-02	0.22749
MAPK12	15	1.04E-01	0.49264
GAPDH	15	1.04E-01	0.49264
PTGS2	12	7.05E-02	0.20295
PDGFB	11	5.26E-02	0.13246

Number of genes per rule = 5 (approximately)
$\begin{matrix} Random occurrence of gene = Number of rules \times \frac{Number of genes per rule}{Total number of genes} \\ = 220 \times \frac{5}{70} = 15.71 \end{matrix}$
Arranged in decreasing order of their frequencies of occurrence in 220 rules, the genes show a general trend towards increasing probability of being selected in a rule by random chance. The postulated frequency of occurrence of each gene if all have equal probabilities of being selected is 15.71. PTGS2 and PDGFB have smaller probabilities than the two genes preceding them because they were actually used less frequently than random chance would suggest.

ISSN: 1471-2407