There seems to be an increasing tend (in both academic and industry) of using methylation pattern for cancer detection / classification. It almost feels that methylation is superior on such task over other omics (somatic mutation or RNA expression). What I can think of this is:
Methylation data at genome level is continous and more 'dense' while mutation data is binary and sparse. Therefore we have more room of feature selection on methylation data. In comparison with RNA-Seq, methylation is advantagous in measurement stability.
On the opposite, methylation is only one factor that influences the expression. Other factors such as all kinds of RNA (microRNA, lncRNA ...) and somatic mutation altogether create a complicated network signal that drive tumorigenesis and it is unclear how methylation weights in contribution among all these factors.
Recent studies, for example from Grail, showed good performance of methylation based prediction on TOO (tissue of origin) as well as classifying more discrete complication such as cancer associated virus infection.
How should I understand that methylation as just one of genotypical feature can achieve such good performance? Can anyone share some discussion on this?