Predicting genome-wide DNA methylation using methylation marks, genomic position, and DNA regulatory elements

Table 1 Performance of methylation status prediction using different prediction models

Feature set	Features	Accuracy (%)	AUC	Specificity (%)	Sensitivity (%)	MCC	R	RMSE
Gene_pos ^a	9	78.6	0.85	72.5	83.6	0.57	0.61	0.39
Gene_pos + seq_property ^b	13	79.5	0.86	71.6	85.9	0.58	0.66	0.34
Gene_pos + seq_property + TFBSs	93	85.7	0.92	78.4	91.7	0.71	0.80	0.29
Gene_pos + seq_property + CREs ^c	118	89	0.94	83.9	93.3	0.78	0.86	0.23
Neighbor CpG methylation status + distance ^d	4	90.7	0.94	87.2	93.5	0.81	0.87	0.24
All features	122	91.9	0.96	87.9	95.1	0.84	0.90	0.19

^aGenomic position features including gene region status (promoter, gene body, and intergenic region), CGI status (CGI, CGI shore, CGI shelf, and non-CGI), and proximal SNPs. ^bDNA sequence properties include GC content, recombination rate, conservation score, and iHSs. ^cCREs include TFBSs, DHS sites, histone modifications and chromatin state segmentation. ^dGenomic distance between neighboring CpG sites. AUC, area under ROC curve; CGI, CpG island; CRE, cis-regulatory element; iHS, integrated haplotype score; MCC, Matthew’s correlation coefficient; RMSE, root-mean-square error; SNP, single nucleotide polymorphism; TFBS, transcription factor binding site.

ISSN: 1474-760X