Prediction of Back-splicing sites for CircRNA formation based on convolutional neural networks

BMC Genomics

Table 2 Comparison of CircCNN and other baseline models in cross-validation

Model	Human		Mouse		Fruit Fly
Model	AUC	ACC	AUC	ACC	AUC	ACC
Model①	0.8614	0.8019	0.8347	0.7669	0.8518	0.7755
Model②	0.7245	0.6744	0.7715	0.7054	0.7716	0.703
Model③	0.8393	0.7793	0.82	0.7525	0.8415	0.7593
Model④	0.8334	0.762	0.7915	0.7188	0.8231	0.7398
Model⑤	0.7117	0.6647	0.7242	0.6637	0.743	0.676
DeepCircCode	0.8827	0.8232	0.8391	0.7653	0.8611	0.7796
CircCNN (CVLD)	0.9026	0.8348	0.8431	0.7572	0.8704	0.7807
CircCNN	0.9049	0.8421	0.8514	0.7705	0.8708	0.7869

CVLD represents the cross-validation strategy used in CircCNN training is same as DeepCirCode

ISSN: 1471-2164