Incorrect predAge warning message: "The number of missing CpGs .. exceeds 80%."

[friedav]

Hello,

Thanks for this great package!

I noticed a potential error in the warning message concerning CpG missingness when running DNAmAge(). As expected, when the CpG missingness exceeds 20%, the epigenetic age is not estimated for a given clock. However, the warning states that more than 80% are missing, which appears to be incorrect. In the case where I noticed this, checkClocks() showed a missingness of 25.7% for the TLclock, but I still received the following warning:

In predAge(cpgs.imp, coefTL, intercept = TRUE, min.perc) :
  The number of missing CpGs forTLclock exceeds 80%.
  ---> This DNAm clock will be NA.

I would suggest to change this to The number of missing CpGs for TLclock exceeds 20%. to avoid potential confusion.

Best,
Friederike

[dpelegri]

Hi Friederike,

I have just pushed the fix to the devel branch.

The warning message now correctly reports the missingness percentage and compares it against the proper threshold, avoiding the confusion between the missing and present CpG counts. For example, if you use the default threshold of 80% and the missingness is higher, the message will now look like this:

Warning: The number of missing CpGs for TL clock is 92.62% which exceeds the threshold of 80%. This DNAm clock will be NA.

Please, could you test it to verify it works as expected on your end?

You can install this updated version using:

if (!require("devtools")) install.packages("devtools")
devtools::install_github("isglobal-brge/methylclock", ref = "devel")

Once you confirm everything is correct, I will proceed with the official submission to the Bioconductor repositories.

Thank you for your patience and for helping us improve methylclock!

Best regards.

[friedav]

Hi Dolors,
does the parameter min.perc with a default value of 80% not reflect the minimum percentage of available CpGs (in contrast to a maximum percentage of missing CpGs)? So in my understanding, if you use the default threshold of 80%, more than 20% missingness will lead to the given clock not being calculated. I.e., I would expect something like this:

warning("The number of available CpGs for ", tit, 
        "clock is below the threshold of ", min.perc*100,
        "%.\n  ---> This DNAm clock will be NA.\n")

or reversely

warning("The number of missing CpGs for ", tit, 
        "clock is above the threshold of ", (1-min.perc)*100,
        "%.\n  ---> This DNAm clock will be NA.\n")

From a user perspective, I personally would find the first option more intuitive, given how the corresponding parameter is coded as min.perc = minimum percentage of available CpGs.

Best,
Friederike

[dpelegri]

Hi Friederike,

First of all, please accept my apologies for the delay in getting back to you. Thank you for the clarification! You are absolutely right. I was focusing on the 'missingness' logic in the code, but since the parameter is named min.perc (minimum percentage of available CpGs), the warning should definitely reflect that perspective to be intuitive for the user.

I agree that your first option is the clearest. I have updated the code to focus on availability. I've changed the logic to report the percentage of available CpGs instead of 'missingness' or 'number', so it aligns perfectly with the min.perc parameter. The message now looks like this:

Warning: The percentage of available CpGs for TL is 74.3%, which is below the threshold of 80%. This clock will be NA.

I have pushed this change to the devel branch. Please let me know if it looks good to you now. Thanks again for your patience and for helping me make methylclock more user-friendly!

Best regards,
Dolors

[friedav]

Hi Dolors,

Thanks again for looking into this! I just had a quick look (sorry for the delay) and overall the warning looks good now. There just seems to be a small issue left regarding the presence of the intercept weight in coefs$CpGmarker for some clocks, leading to an offset of 1 in the calculation of available CpGs.

For example, for the TL clock I get in my target data:

• 31 out of 140 missing CpGs (22.1%) from checkClocks
• The percentage of available CpGs for TL clock is 77.3% which is below the threshold of 80%. ---> This clock will be NA., i.e. 22.7% missing CpGs, from DNAmAge

It looks like there are 140 CpGs plus intercept in the TL clock. In the checkClocks function, the element containing the intercept is excluded before checking the overlap with the target data. In the predAge function, the element seems to be kept when preparing the mask object, leading to an additional incorrectly reported missing CpG. I.e., 32/141 CpGs missing = 22.7%, instead of the expected 31/140 CpGs = 22.1% missing.

Best regards,
Friederike