Table 7.

Performance comparison of the proposed AL6mA and LA6mA methods with other existing methods for predicting 6mA sites on the test datasets

Dataset	Method	Sen1	Spe1	Acc1	MCC1	AUROC	Sen2	Sen3
Arabidopsis thaliana	DeepM6Aa	0.894	0.931	0.913	0.826	0.966	0.920	0.956
	i6mA-DNCb	0.846	0.909	0.878	0.757	0.944	0.853	0.912
	iDNA6mAc	0.843	0.889	0.866	0.733	0.932	0.833	0.902
	3-mer-LRd	0.669	0.728	0.699	0.397	0.773	0.411	0.577
	LA6mA	0.899	0.917	0.909	0.817	0.962	0.912	0.948
	AL6mA	0.862	0.905	0.884	0.768	0.945	0.867	0.927
Drosophila melanogaster	DeepM6Aa	0.901	0.939	0.920	0.841	0.969	0.930	0.959
	i6mA-DNCb	0.869	0.917	0.893	0.787	0.947	0.878	0.916
	iDNA6mAc	0.883	0.843	0.863	0.727	0.937	0.846	0.904
	3-mer-LRd	0.680	0.702	0.691	0.383	0.753	0.347	0.558
	LA6mA	0.909	0.915	0.912	0.824	0.966	0.921	0.955
	AL6mA	0.840	0.916	0.878	0.758	0.941	0.848	0.920

^aResult obtained by retraining and retesting the source code of DeepM6A [28].

^bResult obtained by the re-implementation of i6mA-DNC [30] on benchmark datasets.

^cResult obtained by the re-implementation of iDNA6mA [26] on benchmark datasets.

^dResult obtained by testing the well-trained 3-mer-based LR model.

¹With prediction cutoff threshold value set as 0.5.

²With the fixed Specificity at 0.9.

³With the fixed Specificity at 0.8.