Table 2.

Evaluation of the induced rules using the binding data from Lee et al. (2002)

			Binding data evaluation (significant fractions P < 0.01)
				Random tests
Expression data	Expression similarity thresholds	No. rules unique/all	Rules (P-value)	Similar expression	Common motifs	Random
Cell cycle	0.250	39/109	0.54 (0.000)	0.11	0.17	0.02
Sporulation	0.250	45/81	0.13 (0.708)	0.09	0.18	0.02
Diauxic shift	0.200	150/428	0.29 (0.000)	0.06	0.18	0.02
Heat and cold shock	0.125	52/123	0.52 (0.000)	0.18	0.18	0.02
Pheromone	0.150	53/91	0.39 (0.001)	0.14	0.17	0.02
DNA-damaging agents	0.200	59/116	0.35 (0.000)	0.10	0.17	0.02

A rule is said to be significant if at least one transcription factor binding any of the matching genes obtained a Bonferroni-corrected P-value of <0.01 (only experimental bindings at P < 0.01 from Lee et al. 2002 were considered). The table gives the fraction of significant rules for each data set, and compares these values to what is observed when randomly selecting corresponding sets of genes with only similar expression, common binding sites, or neither. All three random tests produce a P-value that is the probability of observing a higher value than the one reported for the rules. We show the highest of these P-values in parentheses and mark the corresponding random test in bold if this P-value is >0. The table also gives the Euclidean distance threshold (normalized by the number of measurement points) used to define similar expression profiles and the number of rules induced for each expression data set (number of rules unique to that data set/all rules derived for that data set). Additional statistics on the standard deviation of the random test scores and comparisons to a new binding data set (Harbison et al. 2004) may be found at our Web site.