Profile analysis and prediction of tissue-specific CpG island methylation classes

BMC Bioinformatics

Table 10 Comparison of accuracy using binary methylation classification.

Methylation classification	Dataset	Methods	*Acc*[%]	CC
binary	HEP¹	EpiGRAPH – SVM linear kernel*	84.90	0.657
binary	HEP¹	EpiGRAPH – Decision tree C4.5*	75.80	0.462
binary	HEP¹	Matlab – Decision tree**	90.08	0.743
binary	EpiGRAPH ²	EpiGRAPH – SVM linear kernel*	85.20	0.658
binary	EpiGRAPH ²	EpiGRAPH – Decision tree C4.5*	78.60	0.524
binary	EpiGRAPH ²	Matlab – Decision tree**	91.67	0.775
four classes	HEP³	Matlab – Decision tree**	89.39	0.707

*Validation was performed using 10 repetitions of 10 fold cross-validation
** Validation was performed using 10 fold cross-validation
¹HEP CGI data (using our attributes and binary methylation classes)
²EpiGRAPH methylation data (using the default EpiGRAPH sequence attributes and binary methylation classes)
³HEP CGI data (using our attributes and four methylation classes)
The average accuracy (Acc) and correlation coefficient (CC) were used to measure fitness.

ISSN: 1471-2105