Deep learning improves the ability of sgRNA off-target propensity prediction

BMC Bioinformatics

Table 2 The prediction results in Model Selection

Test Set	Model	Recall	ROC_AUC	PRC_AUC
Total test set	CnnCrispr	0.857	0.975	0.679
	CnnCrispr_NoLSTM	0.611	0.987	0.651
	CnnCrispr_Conv_LSTM	0.643	0.986	0.67
	CnnCrispr_NoBatchNor	–	0.5	0.504
	CnnCrispr_NoDropout	0.810	0.985	0.625
Hek293t test set	CnnCrispr	0.864	0.971	0.686
	CnnCrispr_NoLSTM	0.631	0.988	0.658
	CnnCrispr_Conv_LSTM	0.660	0.988	0.694
	CnnCrispr_NoBatchNor	–	0.5	0.504
	CnnCrispr_NoDropout	0.816	0.988	0.636
K562 test set	CnnCrispr	0.826	0.995	0.688
	CnnCrispr_NoLSTM	0.522	0.985	0.589
	CnnCrispr_Conv_LSTM	0.565	0.981	0.57
	CnnCrispr_NoBatchNor	–	0.5	0.503
	CnnCrispr_NoDropout	0.783	0.973	0.597

CnnCrispr had the best comprehensive performance in the three test sets, and the calculated recall value was higher than other pre-selected models. The numbers in boldface indicate the highest scores for each metric

ISSN: 1471-2105