Table 3.

Glimmer3 prediction accuracy with long-orfs training

Genome			Glimmer3 Predictions					versus Glimmer2.13
Organism	GC%	# Genes	3′ Matches		5′ & 3′ Matches		Extra	3′ Match	5′ & 3′	Extra
A.fulgidus	49	1165	1161	99.7%	873	74.9%	1332	−2	−34	−64
B.anthracis	35	3132	3125	99.8%	2751	87.8%	2419	−1	+752	−144
B.subtilis	44	1576	1562	99.1%	1391	88.3%	3020	+3	+421	−724
C.tepidum	57	1292	1289	99.8%	934	72.3%	835	+3	+26	−400
C.perfringens	29	1504	1501	99.8%	1383	92.0%	1192	−1	+267	−20
E.coli	51	3603	3534	98.1%	3112	86.4%	1002	+11	+784	−843
G.sulfurreducens	61	2351	2337	99.4%	1933	82.2%	1165	+7	+575	−734
H.pylori	39	915	910	99.5%	795	86.9%	788	+2	+57	−103
P.fluorescens	63	4535	4510	99.4%	3598	79.3%	1953	+35	+895	−2359
R.solanacearum	67	2512	2485	98.9%	2028	80.7%	1183	+341	+1044	−2184
S.epidermidis	32	1650	1646	99.8%	1514	91.8%	791	+8	+358	−32
T.pallidum	53	575	567	98.6%	391	68.0%	567	−2	+50	−281
U.parvum	26	327	324	99.1%	295	90.2%	297	−1	+21	−11
Averages:				99.3%		83.1%		+31	+401	−608

Genomes and columns are as in the preceding table. Glimmer3 was run by using the output of its long-orfs program to train an IMM. The output of an initial run of Glimmer3 was used to set start codon frequencies and to find a ribosome-binding-site motif. A second run of Glimmer3 using those values generated the above predictions. Glimmer2 was trained on the output of its version of the long-orfs program.