Genome-wide promoter extraction and analysis in human, mouse, and rat

Table 2 Sensitivity and specificity of promoter prediction with different methods

	Sn	Sp
(a) 13,313 unique TSSs in 8,949 human genes
Method 0*	72%	46%
Method 1^†	67%	46%
Method 2^‡	70%	57%
Method 3^§	69%	66%
(b) 9,806 TSSs of 500 bp apart in 8,949 human genes
Method 1 + script^¶	64%	33%
Method 2 + script	67%	44%
Method 3 + script	66%	60%
(c) 6,356 TSSs of 500 bp apart in 5,893 human genes with homologous promoters
Method 1 + script	80%	37%
Method 2 + script	84%	46%
Method 3 + script	82%	69%

*Method 0 used original FirstEF alone to predict promoters in the upstream and genic regions of these genes. ^†Method 1 used de novo FirstEF to predict promoters in the upstream and genic regions of these genes. ^‡Method 2 compared mRNAs or predicted transcripts with original FirstEF predictions to filter out promoters that were neither located in the upstream of the gene region nor overlapping with the 5'-end of any transcripts of this gene. ^§Method 3 tried to first find the promoters in one gene that have homologous rodent promoters. If no such promoters were found, it used Method 2 to select promoters for this gene. ^¶script, a post-clustering script to select representative TSSs from the output of each method described above that were at least 500 bp apart (see Materials and methods for details).

ISSN: 1474-760X