Table 7. Transfer learning performance comparison of models with different sequence lengths on the cervical cancer dataset.

The transfer learning performance of different models using 300 and 600 bp sequence lengths on the cervical cancer dataset. The table includes various evaluation metrics such as loss, accuracy, AUC, F1-score, specificity, and MCC. The results suggest that DNABERT-2_CA_BL performs the best across most metrics, especially with 600 bp sequences. The best results for sequence lengths of 300 bp and 600 bp are indicated in bold.

Length	Model	Loss	Acc	AUC	F1	Sp	MCC
300 bp	DNABERT-2_BASE	0.313	0.872	0.931	0.872	0.827	0.747
	DNABERT-2_CNN	0.372	0.873	0.933	0.873	0.891	0.746
	DNABERT-2_BiLSTM	0.379	0.884	0.874	0.884	0.950	0.775
	DNABERT-2_C M _BL	0.309	0.899	0.942	0.899	0.931	0.799
	DNABERT-2_C A _BL	0.290	0.894	0.949	0.894	0.901	0.789
600 bp	DNABERT-2_BASE	0.362	0.885	0.931	0.885	0.868	0.771
	DNABERT-2_CNN	0.472	0.889	0.939	0.889	0.868	0.778
	DNABERT-2_BiLSTM	0.470	0.882	0.900	0.882	0.881	0.764
	DNABERT-2_C M _BL	0.315	0.895	0.949	0.895	0.908	0.791
	DNABERT-2_C A _BL	0.310	0.893	0.948	0.893	0.881	0.786