how to handle infinite M-values in methylation array data

Question

[matt192]

I'm analyzing EPIC array data. This is my first time and I'm unsure how to deal with infinite M-values. I tested various preprocessing methods. Some produce infinite M-values (caused by red or green intensity values of 0). I used detectionP() to get rid of some of them. But some remain.

Are they meaningful, indicating very low levels of methylation? Or are they artifacts that should be removed?

I'm hoping for feedback from someone who is experienced in analyzing methylation arrays. Thanks in advance!

Here are the number of infinite M-values for each processing method:

raw illumina     swan  funnorm quantile     noob 
878    44575        0       23        0        0

Answer 1

[Basti]

This is a known problem happening with some of the normalization methods. As long as the detection P-value is ok, you can keep these probes as they are meaningful.

Depending on the normalization you will use, you will indeed encounter this issue. If you finally choose a normalization with this problem happening, you could replace all -Inf values with the minimum M-value of your dataset, and inversely for the +Inf values.

Alternatively, you could use the shiftBeta function from the Harman package, which is doing slightly the same thing by replacing infinite values with a chosen value (near 0.0001 or 0.9999 for instance) : https://rdrr.io/bioc/Harman/man/shiftBetas.html

This is almost similar to what is done with noob normalization for instance as you can see in the source code that they replace 0 Meth and Unmeth signals with 1 instead.

Meth[Meth <= 0] <- 1L
Unmeth[Unmeth <= 0] <- 1L