Comparative Analysis on Alignment-Based and Pretrained Feature Representations for the Identification of DNA-Binding Proteins

Table 3

Results of embedded FS methods using three regularizers of the linear model in the light of 5CV on PDB1616.

Feature set	FS method	#Features	Fivefold cross-validation on PDB1616
			ACC (%)	MCC	SP (%)	SN (%)
PSSMR_All	NFS	580	71.47	0.4306	67.82	75.12
	ElasticNet	188	72.77	0.4563	69.68	75.87
	Lasso	61	72.83	0.4577	69.43	76.24
	LassoLars	58	73.51	0.4707	71.53	75.50
PSSMS_All	NFS	580	72.77	0.4597	65.97	79.58
	ElasticNet	207	74.01	0.4847	67.20	80.82
	Lasso	54	73.82	0.4794	68.32	79.33
	LassoLars	38	72.77	0.4586	66.96	78.59
ESM_Avg	NFS	1280	79.27	0.5908	72.52	86.01
	ElasticNet	430	81.93	0.6442	75.37	88.49
	Lasso	142	83.11	0.6656	77.97	88.24
	LassoLars	151	82.43	0.6514	77.72	87.13
ESM_All	NFS	37120	78.90	0.5843	71.53	86.26
	ElasticNet	884	86.14	0.7267	80.94	91.34
	Lasso	367	87.87	0.7598	83.91	91.83
	LassoLars	250	86.70	0.7353	83.66	89.73