TABLE 3.

Performance of three ensemble machine learning-based classifiers for circRNA regulatory interactions prediction based on different groups of sequence-derived features.

Features	ROC	SE	SP	ACC	MCC	F1	PRE
(A) circRNA-miRNA classifier
Sequence-based	0.995	1.000	0.991	0.995	0.990	0.995	0.990
Graph features	0.958	0.980	0.937	0.957	0.915	0.956	0.933
Genome context	0.884	0.876	0.892	0.883	0.766	0.889	0.904
Regulation information	0.952	0.922	0.981	0.952	0.906	0.950	0.979
(B) circRNA-RBP classifier
Sequence-based	0.991	0.986	0.995	0.990	0.981	0.991	0.995
Graph features	0.969	0.954	0.985	0.969	0.938	0.970	0.986
Genome context	0.895	0.833	0.956	0.894	0.794	0.888	0.951
Regulation information	0.955	0.925	0.985	0.954	0.910	0.954	0.985
(C) circRNA-TR classifier
Sequence-based	0.993	0.989	0.998	0.993	0.986	0.993	0.998
Graph features	0.961	0.935	0.987	0.962	0.925	0.959	0.985
Genome context	0.891	0.954	0.829	0.891	0.788	0.896	0.846
Regulation information	0.971	0.978	0.965	0.971	0.943	0.970	0.962