Table 1.

Performance comparison between EnDeep4mC ensemble model and each base model trained by fivefold cross-validation on six species data sets

Data set	Algorithm	ACC	SN	SP	MCC	AUC	F1-score
C. elegans	CNN	0.9252	0.9559	0.8946	0.8529	0.9812	0.9278
	Bi-LSTM	0.9231	0.9514	0.8948	0.8461	0.9792	0.9246
	Transformer	0.9296	0.9589	0.9003	0.859	0.9829	0.9308
	EnDeep4mC	0.9571	0.9594	0.9548	0.9142	0.9914	0.9572
D. melanogaster	CNN	0.9191	0.9530	0.8851	0.8401	0.9758	0.9216
	Bi-LSTM	0.9169	0.9502	0.8836	0.8379	0.9738	0.9205
	Transformer	0.9246	0.9540	0.8951	0.8512	0.9778	0.927
	EnDeep4mC	0.9412	0.9508	0.9316	0.8826	0.9842	0.9418
A. thaliana	CNN	0.8727	0.8933	0.8522	0.7489	0.9453	0.8758
	Bi-LSTM	0.8702	0.8781	0.8623	0.7423	0.9423	0.8727
	Transformer	0.8789	0.8962	0.8617	0.7587	0.9491	0.8811
	EnDeep4mC	0.9133	0.9205	0.9061	0.8267	0.9697	0.9133
E. coli	CNN	0.9619	0.9513	0.9725	0.9438	0.9933	0.9718
	Bi-LSTM	0.9644	0.9437	0.9850	0.9251	0.9972	0.9623
	Transformer	0.9675	0.9613	0.9737	0.9376	0.9947	0.9686
	EnDeep4mC	0.9973	0.9969	0.9976	0.9945	0.9999	0.9973
G. subterraneus	CNN	0.8525	0.8725	0.8325	0.7102	0.9334	0.852
	Bi-LSTM	0.8551	0.8545	0.8557	0.7142	0.9309	0.8578
	Transformer	0.8514	0.8527	0.8500	0.7024	0.9282	0.8513
	EnDeep4mC	0.9349	0.9323	0.9375	0.8698	0.9786	0.9347
G. pickeringii	CNN	0.9075	0.9437	0.8712	0.8233	0.9722	0.9138
	Bi-LSTM	0.9180	0.9384	0.8976	0.8345	0.9742	0.9186
	Transformer	0.9107	0.9346	0.8868	0.8241	0.9709	0.9127
	EnDeep4mC	0.9729	0.9740	0.9718	0.9457	0.9937	0.9729

Bold values indicate the best performance achieved within each comparison group.

EnDeep4mC predicts DNA N⁴-methylcytosine sites using a dual-adaptive feature encoding framework in deep ensembles

EnDeep4mC predicts DNA N4-methylcytosine sites using a dual-adaptive feature encoding framework in deep ensembles