Comparative analysis and prediction of nucleosome positioning using integrative feature representation and machine learning algorithms

Table 2 The prediction results of integrative feature representation for H. sapiens via 10-fold cross-validation by SVM, ELM and XGBoost

Method	Feature	parameter	ACC	Sn	Sp	MCC	AUC
SVM	FCGR + DAC	K = 4, lag = 2	0.8708	0.8896	0.8522	0.7425	0.9315
	FCGR + TAC	K = 4, lag = 2	0.8679	0.8878	0.8483	0.7369	0.9288
	FCGR + DACC	K = 4, lag = 2	0.8537	0.8531	0.8544	0.7079	0.9208
	FCGR + TACC	K = 4, lag = 2	0.8415	0.8319	0.8509	0.6837	0.9113
	FCGR + PCPseDNC	K = 4, λ = 8, w = 0.5	0.8673	0.8936	0.8413	0.7359	0.9286
	FCGR + PCPseTNC	K = 4, λ = 8, w = 0.5	0.8708	0.8966	0.8452	0.7429	0.9273
	All features		0.8137	0.7518	0.8748	0.6322	0.8996
ELM	FCGR + DAC	K = 4, lag = 2	0.8292	0.8539	0.8048	0.6598	0.9007
	FCGR + TAC	K = 4, lag = 2	0.8297	0.8531	0.8065	0.6604	0.8977
	FCGR + DACC	K = 4, lag = 2	0.8336	0.8627	0.8048	0.6689	0.9009
	FCGR + TACC	K = 4, lag = 2	0.8325	0.8632	0.8022	0.6668	0.8983
	FCGR + PCPseDNC	K = 4, λ = 8, w = 0.5	0.8314	0.8658	0.7974	0.6648	0.8985
	FCGR + PCPseTNC	K = 4, λ = 8, w = 0.5	0.8248	0.8544	0.7957	0.6516	0.8947
	All features		0.8356	0.8632	0.8083	0.6735	0.9013
XGBoost	FCGR	K = 1 + 2 + 4	0.8585	0.89309	0.8244	0.71934	0.9197
	FCGR + DAC	K = 4, lag = 2	0.8450	0.87503	0.8152	0.69182	0.9160
	FCGR + TAC	K = 4, lag = 2	0.8402	0.8733	0.8074	0.68221	0.9136
	FCGR + DACC	K = 4, lag = 2	0.8423	0.86583	0.8191	0.68588	0.9127
	FCGR + TACC	K = 4, lag = 2	0.8391	0.87287	0.8057	0.68059	0.9115
	FCGR + PCPseDNC	K = 4, λ = 8, w = 0.5	0.8559	0.88913	0.8230	0.71396	0.9207
	FCGR + PCPseTNC	K = 4, λ = 8, w = 0.5	0.8498	0.88254	0.8174	0.70168	0.9183
	All features		0.8472	0.87374	0.8209	0.69581	0.9170

All features means the feature vector = FCGR + DACC + TACC + PC-PseDNC + PC-pseTAC, and the parameters are consistent with the parameters of the corresponding feature. Parameter K indicates the values of K nucleotide in FCGR; lag indicates the distance of lag along the sequence; λ represents the highest counted rank (or tier) of the correlation along a DNA sequence; w is the weight factor ranged from 0 to 1
Best values are in bold

ISSN: 1471-2105