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

BMC Bioinformatics

Table 1 Evaluation results in K562 cells, based on annotated protein-coding gene TSSs. From left to right, the number of total positive predictions of each algorithm is shown, the number of TPs and FPs in both CAGE- and gene-oriented benchmarks as well as the performance in terms of precision and sensitivity on default parameters

Algorithm	Total positive predictions	All predictions in query zone		Gene − oriented set of predictions		Protein-coding TSS annotation
Algorithm	Total positive predictions	TP	FP	TP	FP	Precision	Sensitivity
DeepTSS	31,443	6398	123	3122	91	0.98	0.96
ADAPT-CAGE	31,177	6294	172	3091	125	0.97	0.94
CAGER	14,465	6489	1771	3102	1003	0.97	0.78
PARACLU	9453	4016	129	2258	95	0.97	0.60
RECLU	11,558	6257	1649	3082	970	0.93	0.79
TOMETOOLS	30,689	5765	228	3016	174	0.96	0.86
iTiSS	1734	98	37	118	40	0.72	0.01

ISSN: 1471-2105