Construction and use of gene expression covariation matrix

BMC Bioinformatics

Table 2 Statistical properties of networks

chipset	network	#biol	#comp	#raw	#sel	%sel	mco	mc	mdc	ma	mda
Human Genome	ALL1	71	2485	74017943	8400736	11.35	1331	36.21	22.38	13.31	13.86
U95 Set (12 6245 ps)	ALL10			78414854	16751383	21.36	2654	26.81	22.85	15.95	17.10
	NR1	37	666	70749360	2291090	3.24	363	42.10	20.86	29.07	21.03
	NR10			77866285	14260641	18.31	2259	29.63	23.43	20.01	20.91
Human Genome	ALL1	126	7875	182604317	7963686	4.36	715	35.71	23.98	9.10	8.81
U133 Set (22 283 ps)	ALL10			235733279	20713103	8.79	1859	29.37	22.47	11.90	12.27
	NR1	88	3828	160751542	3704369	2.30	332	47.68	30.04	9.12	8.25
	NR10			227849133	10331556	4.53	927	38.04	26.40	11.32	11.29
Murine Genome	ALL1	205	20910	75925711	11013118	14.51	1764	22.37	18.48	11.87	12.26
U74 Version 2 (12 488 ps)	ALL10			76566772	40596332	53.02	6502	20.63	21.28	18.71	20.37
	NR1	89	3916	74613508	2567219	3.44	411	38.18	28.10	11.91	12.12
	NR10			77231972	4633426	6,00	742	33.24	27.24	15.52	15.98
Rat Genome	ALL1	96	4560	25130304	2645806	10.53	601	36.69	26.01	16.07	16.64
U34 Set (8 799 ps)	ALL10			35636045	6508820	18.26	1479	29.99	23.92	16.69	17.58
	NR1	41	820	19937680	902388	4.53	205	55.05	46.33	23.35	23.26
	NR10			33301044	1801457	5.41	409	41.68	32.50	21.72	22.19

For each of the models listed, four networks were constructed based upon the combination of two independent factors: either all the biological conditions retained at the end of the two selection steps or only the non-redundant ones (labelled ALL and NR, respectively), and either 1% or 10% FDR (labelled 1 and 10, respectively). For each network, the table displays the number of biological conditions (#biol), the number of comparisons performed (#comp), the number of positive correlation scores in CORR (#raw), the number and percentage of statistically significant positive correlation scores (#sel, %s), the mean connectivity of a node (mco), and the mean and median positive (mc and mdc) or negative (ma and mda) correlation values linking a node to its neighbours. The values in the last five columns were computed on statistically significant correlation scores. The #raw, #sel and %sel values for the negative correlation scores in ANTI were similar to those of CORR and are therefore not indicated.

ISSN: 1471-2105