Enhancer prediction in the human genome by probabilistic modelling of the chromatin feature patterns

BMC Bioinformatics

Table 3 The AUC values for the genome-wide predictions

Cell line	method	#enhancer predictions	p300	TRF
K562	PREPRINT ML	35089	0.857	0.874
K562	PREPRINT Bayesian	35089	0.855	0.887
K562	RFECS	35089	0.911	0.900
K562	PREPRINT ML	15531	0.850	0.898
K562	PREPRINT Bayesian	15531	0.856	0.908
K562	RFECS	15531	0.912	0.929
GM12878	PREPRINT ML	49699	0.826	0.889
GM12878	PREPRINT Bayesian	49699	0.833	0.888
GM12878	RFECS	49699	0.911	0.895
GM12878	PREPRINT ML	22088	0.821	0.911
GM12878	PREPRINT Bayesian	22088	0.826	0.910
GM12878	RFECS	22088	0.881	0.941

The true labels of the predictions were based on the overlap between the predictions and the TRF ChIP-seq peaks. An equal number of enhancers predicted by PREPRINT and RFECS were chosen; the numbers are indicated in the column #enhancer predictions. In each combination of the validation data (p300 or TF) and the method, the AUC value for the best method was highlighted with the bold font. In addition, the AUC values for PREPRINT predictions in cell line GM12878, computed utilising the TRF validation data, were highlighted due to a comparable generalisation performance to RFECS

ISSN: 1471-2105