A new profiling approach for DNA sequences based on the nucleotides' physicochemical features for accurate analysis of SARS-CoV-2 genomes

Table I Investigating the encoding capability of the PC-mer (using only Linear SVM classifier) vs. that of the FCGR (using six classifiers: Linear Discriminant (LD), Linear SVM (LSVM), Quadratic SVM (QSVM), Fine KNN (FKNN), Subspace Discriminant (SD), and Subspace KNN (SKNN)) as the encoding methods for generating input vectors

Encoding algorithm	k-mer	Classification model	Metrics	Datasets (%)
Encoding algorithm	k-mer	Classification model	Metrics	*Test-1*	*Test-2*	*Test-3a*	*Test-3b*	*Test-4*	*Test-5*	*Test-6*	*Human Coronavirus*
PC-mer	12	Linear SVM	Accuracy	97.25	95.93	98.52	100	100	99.33	100	100
			F1	97.23	95.9	98.49	100	100	99.36	100	100
			Precision	97.38	96.16	98.85	100	100	99.55	100	100
			Recall	97.25	95.93	98.52	100	100	99.33	100	100
FCGR	7	LD	Accuracy	91.7	91.2	98.1	100	97.6	98.6	100	-
		LSVM		90.8	89.2	94.2	93.3	98.4	97.4	100	-
		QSVM		95	93.1	95.2	93.3	98.4	97.4	100	-
		FKNN		93.4	90.3	95.7	95	97.6	97.4	100	-
		SD		87.6	89	97.6	95	98.4	98.7	100	-
		SKNN		93.2	90.4	96.2	95	97.2	96.1	100	-
		Average accuracy		92	90.5	96.2	95.3	97.6	97.5	100	-

ISSN: 1471-2164