Sure, just do the PCA or correlation on the logit transformed ratio. You could also try this on the ratio (i.e., the percent methylation), but I'm not sure how the assymetric variance will affect things.

Principal components can be calculated for any numeric matrix. Question, I guess, is what you want to extract from PCA. Usually PCA is used either as an exploratory analysis to see if samples (or loci) tend to cluster together depending, in your case, on depth of coverage or %methylation. Otherwise PCA can be used to reduce the dimensionality of your dataset, e.g. instead of thousands of features (e.g. CpG) you work with the first few principal components which might summarize well the variability of the whole dataset.