Table 2.

List of Currently Available Tools for 4mC Sites Prediction Assessed in This Study

Year	Toola	Classifier	Training/Independent Dataset Size	Features	Web Server	Evaluation Strategy	File Upload
2017	iDNA4mCb	SVM	Chen dataset/–	RFHCP	yes	LOOCV	no
2018	4mCPredb	SVM	Chen dataset/–	PSTNP, EIIP	yes	LOOCV	no
	4mcPred-SVMb	SVM	Chen dataset/–	Kmer, MBE, DBE, LPDF	yes	10-fold CV	yes
2019	Meta-4mCpredb	RF, ERT, GB, SVM	Chen dataset/Manavalan dataset	Kmer, MBE, DPE, LPDF, RFHCP, DPCP, TPCP	yes	10-fold CV	yes
	4mcPred-IFLb	SVM	Chen dataset/–	Kmer+MBE, DBE+LPDF, PCPs, PseDNC, KNN, EIIP, MMI, RFHCP	yes	10-fold CV	yes
	4mCCNNb	CNN	Chen dataset/–	MBE	yesc	10-fold CV	–
	4mCpred-ELd	RF, GB, ERT, SVM	(800 4mCs and 800 non-4mCs)/(180 4mCs and 180 non-4mCs)	Kmer, DPE+LPDF, RFHC, EIIP, MBE, DPCP, TPCP	yes	10-fold CV	yes
	i4mC-ROSEe	RF	(4854 4mCs and 4854 non-4mCs)/(1617 4mCs and 1617 non-4mCs)	KSNC, MBE, EIIP	yes	10-fold CV	yes
2020	iEC4mC-SVMd	SVM	(388 4mCs and 388 non-4mCs)/(134 4mCs and 134 non-4mCs)	MBE, RFHC, DAE, X-k-YCF, Kmer	–	10-fold CV	–
	DNA4mC-LIPd	–	–/Manavalan dataset	integration of six existing predictors	yes	independent evaluation	yes
	4mcDeep-CBId	CNN, BLSTM	(1,173 4mCs and 6,635 non-4mCs)/ –	same as used in 4mcPred-IFL	–	3-fold CV	–
	iDNA-MSf	RF	7,899 samples/7,898 samples	Kmer, RFHCP, MBE	yes	5-fold CV	yes
	i4mC-Moused	RF	(746 4mCs and 746 non-4mCs)/(160 4mCs and 160 non-4mCs)	Kmer, KSNC, MBE, EIIP	yes	10-fold CV	yes

Chen dataset contains C. elegans (4mCs, 1,554; non-4mCs, 1,554), D. melanogaster (4mCs, 1,769; non-4mCs, 1,769), A. thaliana (4mCs, 1,978; non-4mCs, 1,978), E. coli (4mCs, 388; non-4mCs, 388), Geoa. subterraneus (4mCs, 906; non-4mCs, 906), and Geob. pickeringii (4mCs, 569; non-4mCs, 569). Manavalan dataset contains C. elegans (4mCs, 750; non-4mCs, 750), D. melanogaster (4mCs, 1,000; non-4mCs, 1,000), A. thaliana (4mCs, 1,250; non-4mCs, 1,250), E. coli (4mCs, 134; non-4mCs, 134), Geoa. subterraneus (4mCs, 350; non-4mCs, 350), and Geob. pickeringii (4mCs, 200; non-4mCs, 200). SVM, support vector machine; RF, random forest; GB, gradient boosting; CNN, convolutional neural network; BLSTM, bidirectional long short-term memory network; ERT, extremely randomized tree; RFHCP, ring-function-hydrogen-chemical properties, PSTNP, position-specific trinucleotide propensity; EIIP, electron-ion interaction pseudopotential; Kmer, Kmer nucleotide frequency; MBE, mononucleotide binary encoding, DBE, dinucleotide binary encoding, LPDF, local position-specific dinucleotide frequency; DPE, dinucleotide binary profile encoding; DPCP, dinucleotide physicochemical properties; TPCP, trinucleotide physicochemical properties; PCP, physicochemical property; PseDNC, pseudo-dinucleotide composition; KNN, K-nearest neighbor; KSNC, k-space nucleotide composition; DAC, dinucleotide physicochemical properties autocorrelation; X-k-YCF, Xmer-kGap-Ymer composition frequency; ANF, accumulated nucleotide frequency; LOOCV, leave-one-out cross-validation; CV, cross-validation.

^aThe listed tool URL addresses are as follows: iDNA4mC, http://lin-group.cn/server/iDNA4mC/; 4mCPred, http://server.malab.cn/4mCPred/; 4mcPred-SVM, http://server.malab.cn/4mcPred-SVM/; Meta-4mCpred, http://thegleelab.org/Meta-4mCpred/; 4mcPred-IFL, http://server.malab.cn/4mcPred-IFL/; 4mCCNN, https://home.jbnu.ac.kr/NSCL/4mCCNN.htm; 4mCpred-EL, http://thegleelab.org/4mCpred-EL/; i4mC-ROSE, http://kurata14.bio.kyutech.ac.jp/i4mC-ROSE/; DNA4mC-LIP, http://i.uestc.edu.cn/DNA4mC-LIP/; iDNA-MS, http://lin-group.cn/server/iDNA-MS/; i4mC-Mouse, http://kurata14.bio.kyutech.ac.jp/i4mC-Mouse/.

^bTools contain six species-specific prediction models, namely A. thaliana, C. elegans, D. melanogaster, E. coli, Geoa. subterraneus, and Geob. pickeringii.

^cWeb server is not functional.

^dTool contains one prediction model to compute 4mC site from specific species.

^eTool contains two prediction models for F. vesca and Rosa chinensis.

^fTool contains four different species-specific models, namely F. vesca, Casuarina equisetifolia, Saccharomyces cerevisiae, and Ts. SUP5-1.