I have Infinium methylation EPIC data. I am experimenting with different normalisation to figure out which is the best. All samples come from the same tissue which is blood. Two metrics that I thought would be useful are variance between samples for each probe and sample to sample correlation.
Fig: Left shows boxplot of variance between samples for each probe using different normalisations. Right shows boxplot of sample-sample correlations for various normalisations.
It seems like quantile gives the lowest median probe variance and the tightest sample correlation. Although noob and func shows higher median sample correlation. Another concern is whether quantile might be over correcting and removing useful variation. I would like to know if anyone has any thoughts on this.
I have created MDS plots from these different normalisations which is in this question.