Table 2

Selection of bootstrap cutoff values for plant-specific HMM training sets. Each row shows the results of building a new HMM using 80 initial training sequences plus the number of supplementary bootstrap sequences shown. "Inclusion threshold" indicates minimum score of the bootstrap sequences using HMM80. Accuracy and coverage were determined using the same positive and negative test set for each HMM. Jack-knife (leave-one-out) testing for each HMM was performed against the same training set used in model construction.

Model Name
Threshold Cutoff
Pattern length
Number Bootstrap Sequences
Inclusion Threshold
Accuracy (TP+TN)/TOTAL
Coverage TP/(TP+FN)
ROC area
Jack-knife Detection

HMM80
2.05
22
0
-
100.0%
100.0%
0.969
82.5%
HMM109
0.85
21
29
> 12.5
100.0%
100.0%
0.969
92.7%
HMM166
0.65
21
86
> 9.6
100.0%
100.0%
0.968
96.4%
HMM185
1.40
20
105
> 8.8
100.0%
100.0%
0.969
96.2%
HMM266
0.55
19
186
> 5.8
100.0%
100.0%
0.973
98.5%
HMM319
1.35
20
239
> 4.4
99.6%
98.8%
0.969
97.5%
HMM366
2.80
18
286
> 3.1
99.2%
97.5%
0.969
91.0%
HMM428
0.50
17
348
> 2.0
98.5%
98.8%
0.968
97.0%

Podell and Gribskov BMC Genomics 2004 5:37   doi:10.1186/1471-2164-5-37