Skip to main content

Table 2 Summarized evaluation results in K562 cells, based on chromatin states, experimental data and annotated protein-coding TSSs. From left to right, each column shows the number of total positive predictions of each algorithm, the percentage overlapping with chromatin states associated with active and weak/repressed transcription, and at least one TF and H3K4me3 peak

From: DeepTSS: multi-branch convolutional neural network for transcription start site identification from CAGE data

Algorithm

Total positive predictions

ChromHMM active transcription

ChromHMM weak transcription

TF ChIP-Seq peaks

H3K4me3 ChIP-Seq peaks

DeepTSS

31,443

96.66%

3.33%

98.27%

92.04%

(30,376)

(1047)

(30,898)

(28,939)

ADAPT-CAGE

31,177

95.90%

4.09%

97.72%

91.12%

(29,885)

(1276)

(30,466)

(28,409)

CAGER

14,465

84.62%

15.37%

87.67%

81.11%

(12,195)

(2215)

(12,681)

(11,732)

PARACLU

9453

94.01%

5.98%

95.26%

92.06%

(8868)

(565)

(9,005)

(8702)

RECLU

11,558

93.35%

6.64%

94.32%

89.96%

(10,773)

(767)

(10,902)

(10,397)

TOMETOOLS

30,689

92.59%

7.40%

95.46%

88.12%

(28,395)

(2272)

(29,296)

(27,044)

iTiSS

1734

41.58%

58.41%

48.12%

41.94%

(721)

(1013)

(848)

(739)