Table 2.

Promoter predictions of different models

Training promoters used in model	PWM regions used in model	Predicted promotersa		Differenceb
		M. jannaschii genome	Randomized genome
In vitro selectionc	BRE/TATAd	5067 ± 2286	4042 ± 1940	1026 ± 348
Protein-coding genese	BRE/TATA	2204 ± 849	1436 ± 665	768 ± 191
Protein-coding genes	Extendedf	1386 ± 561	551 ± 323	835 ± 243

^aThe mean number ± SD of total predicted promoters for 100 random assortments of the 134 mapped protein promoters between a training set (100 or 101) and a testing set (34 or 33). For each assortment, the threshold for each model was set to predict 50% of the 34 or 33 testing set promoters.

^bDifference in number of predicted promoters between the M. jannaschii and randomized genomes.

^cSixty promoters from Li et al. (14). Over the 100 replicates (see footnote a), the PWM of the model did not change, but the threshold was adjusted to detect 50% of the testing set promoters.

^dThe BRE (9 nt) and the TATA box (8 nt) plus 4 nt on each side.

^eThe 100 or 101 promoters in the training set for each of the 100 replicates (see footnote a).

^fBRE/TATA-box, PPE/Inr, and spacer score (see Materials and Methods section).