. 2023 Jun 23;13:10221. doi: 10.1038/s41598-023-34795-4

Table 2.

Overview of the deep learning-based methods for splice site detection available in the literature: (D) Donor, (A) Acceptor, Acc. (Accuracy), Sn. (Sensitivity), Sp. Specificity.

Reference	Base	Architecture	Data	Sequence numbers/lengths		Measure	Performance
DeepSplice⁴³	CNN	Input layer 2 Conv layers 1 Dense layer Output layer	GENCODE	291,030 (true) 271,937 (false)	120 nt		Donor	Acceptor
						Acc.	0.907	0.893
						Sn.	0.917	0.873
						Sp.	0.897	0.913
SpliceRover²⁶	CNN	Input layer 2 conv layers+ Max pooling layer 1 Dense layer Output layer	NN269	1324 (true) (D) 4922 (false)(D) 1324 (true) (A) 5553 (false)(A)	15 nt(D) 90 nt(A)	Acc.	0.9535	0.9612
						Sn.	0.9011	0.9077
						Sp.	0.9674	0.9739
						auPRC	0.9829	0.9899
SpliceFinder²⁸	CNN	Input layer A conv layer Dense layer Output layer	Ensembl (hg38 dataset)	30,000^b	40–400 nt	Acc.	0.969–0.832 (40 nt) 0.965–0.903 (400 nt)	0.969–0.832 (40 nt) 0.965–0.903 (400 nt)
DeepSS⁴⁴	CNN	Input layer 2 Conv layers+ Max pooling layer 2 dense layer Output layer	CE	750 (true)(D) 19,250(false)(D) 1000 (true)(A) 19,000 (false)(A)	141 nt	Acc.	0.97^a	0.96^a
						Sn.	0.95^a	0.93^a
						Sp.	0.97^a	0.96^a
						Pr.	0.87^a	0.83^a
						MCC	0.89^a	0.85^a
						AUC ROC	99.47^a	99.56^a
						AUC PR	97.88^a	98.18^a
DeepSS⁴⁴	CNN	Input layer 2 Conv layers+ Max pooling layer 2 dense layer Output layer	NN269	1324(true)(D) 4922(false)(D) 1324(true)(A) 5553(false)(A)	15 nt(D) 90 nt(A)	Acc.	0.93^a	0.97^a
						Sn.	0.91^a	0.93^a
						Sp.	0.96^a	0.97^a
						Pr.	0.86^a	0.9^a
						MCC	0.85^a	0.9^a
						AUC ROC	98.43^a	99.34^a
						AUC PR	93.97^a	97.32^a
Splice2Deep²⁹	CNN	Input layer Conv. layer Max pooling layer Output layer	Ensembl (hg38 dataset)	250,400 (true) 250,400 (false)	602 nt	Acc.	97.38	96.91
						Sn.	95.93	95.61
						Sp.	98.83	97.8
						F1 Score	96.38	96.91
						AUC	99.1	98.69
Sarkar et. al.³¹	Vanilla RNN, LSTM, GRU	3 stacks (90 Vanilla RNN, GRU or LSTM cells)	GenBank (splice-junction gene sequences)	3175^b	60 nt	Acc.	1.00	99.95
						Sn.	1.00	1.00
						Sp.	1.00	99.93
						F1 Score	1.00	99.93
Splice AI²³	ResNet	Input layer Conv. layer Residual blocks (Batch-normalization layers Rectified linear units Convolutional layers) Output layer	GENCODE	13,796 donor–acceptor pairs	40–5000 nt		Combined donor and acceptor
						Acc.	0.95
						AUC PR	0.98

^aThe approximate values were taken from the graphs since exact values were not given in the paper. Only imbalanced dataset results were mentioned for DeepSS.

^bThe number of true and false splice site sequences was not mentioned.