No normalization for EPIC Illumina arrays with RnBeads?

[Noboru Sakabe]

I am new to methylation analysis and I'm using RnBeads (RnBeads_1.4.0) for the first time.

I'm running the code below and the section "Normalization" in the report says:

"Sample average methylation cannot be visualized because no valid Sentrix ID and Sentrix Position information could be extracted from the sample annotation table."

I'm sure my sample_annotation.csv has the Sentrix ID and position. Looking at the standard output, however, I see this:

2017-02-22 18:19:35     1.7  STATUS         STARTED Normalization Procedure
2017-02-22 18:19:37     1.8 WARNING             Incompatible methods for object and method: normalization with method swan cannot be applied to MethylationEPIC data at the moment. Changed the normalization method to "none"
2017-02-22 18:19:37     1.8 WARNING             Incompatible methods for object and background correction method: ]
                            no background correction on the MethylationEPIC data possible at the moment

I thought RnBeads could be used for EPIC array analysis, but maybe it can't fully analyze EPIC data?

Thanks!


This is my code:

library('RnBeads')
data.dir <- "."
idat.dir <- file.path(data.dir, "idat")
sample.annotation <- file.path(idat.dir, "sample_annotation.csv")

analysis.dir <- "rnbeads/analysis"

rnb.options(filtering.sex.chromosomes.removal=TRUE, identifiers.column="Sample_ID")
rnb.run.analysis(dir.reports='rnbeads/reports', sample.sheet=sample.annotation, data.dir=idat.dir, data.type="idat.dir")

# sample_annotation.csv

Sample_Name,Sample_Well,Sample_Plate,Sample_Group,Pool_ID,Sentrix_ID,Sentrix_Position
S1,A1,1,CORZ-24S-A1,,200673610032,R01C01
S2,B1,1,CORZ-24S-B1,,200673610032,R02C01
S3,C1,1,CORZ-24S-C1,,200673610032,R03C01

[Pavlo Lutsik]

Dear Noboru,

If I got it right, you have faced two independent problems.

First, the specific function in RnBeads for plotting average methylation per sample was not able to recognise the Sentrix_ID and Sentrix_Position columns in your sample sheet. This is not a general problem, otherwise you would not have been able to load your data and reach this stage. Could you let me know which tokens you used to define them?

Second, what concerns EPIC arrays, our complete analysis pipeline can be applied to this type of data as well. However, not all normalisation methods which were earlier developed for 450k and were wrapped by RnBeads can deal with EPIC data, which is mostly due to failures in the third-party implementations we rely upon. Nevertheless, many normalisation methods do work with EPIC (for instance "methylumi.illumina" and "wm.*" ones) so you should give them a try. For background correction we recently added a customised implementation of ENMix (https://www.ncbi.nlm.nih.gov/pubmed/26384415) which also supports EPIC. You may want to try those options and let us know whether it helped. In any case, try to avoid "over-normalizing" your data, especially if your study is on matched samples, longitudinal or twin-based.

Best regards,

Pavlo

[Jean-Philippe Fortin]

Dear Noboru,

Just to let you know that minfi has been fully adapted to the EPIC array; all previous normalization methods previously developed for 450k also work for EPIC arrays (noob background correction, SWAN normalization, Functional normalization and others). If you have full blood samples, you can also perform cell type composition estimation on the EPIC arrays using 450k reference blood panels. Our methods as discussed in a recent paper: https://academic.oup.com/bioinformatics/article/33/4/558/2666344/Preprocessing-normalization-and-integration-of-the

Hope this helps,

JP

[Noboru Sakabe]

Dear Pavlo, thanks for the speedy reply.

The Sentrix error is gone, I don't know what happened.

I tried running BMIQ and ENMIX and had no problems (after upgrading RnBeads), thank you!

I find the html report very cool and it would be helpful for beginners to see the kind of guidance you gave me here in the report.

Thank you for the advice about not over-normalizing. I am going to post a question about your comment in another thread.

[Noboru Sakabe]

Dear Jean-Phillippe,

Thanks for the information on minfi!

I will run minfi as well.

[Allen Yu]

Dear Pavlo and Jean-Philippe,

I am currently using RnBeads 1.7.5 and Bioconductor 3.5. When I tried to use minfi.funnorm to normalise EPIC array data, I met the following error:

Error in .SummarizedExperiment.charbound(i, rownames(x), fmt) :

  <MethylSet>[i,] index out of bounds: cg13869341 cg14008030 ... cg08265308 cg14273923

When I look at normalization.R (line 418-419), it seems like only 450k annotation package is loaded, even though minfi.funnorm is one of the supported normalization methods for EPIC (line 211). Could this be a bug?

Best regards,

Allen

[Jean-Philippe Fortin]

Hi Allen, 

I am only maintaining the preprocessFunnorm function in minfi, which works with EPIC arrays. 

Best,

JP

[Pavlo Lutsik]

Hi Allen,

Thanks for reporting, I will have a closer look at it. Unfortunately, the output of preprocessFunnorm did not fit into the model of our normlization module (at least at the time of implementation it was not possible to get normalized intensities) and we had to use non-exported code. When the latter changes such situations occur.

I will come back to you with a fix/workaround.

Best,

Pavlo

--
You received this message because you are subscribed to the Google Groups "Epigenomics forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to epigenomicsfor...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Allen Yu]

Hi Pavlo,

Thanks for looking into the issue! Looking forward to the fix / workaround.

Allen

[invivogen]

Dear Pavlo,

Dear Allen,

Wa there a fix/workaround for this problem finally?

Thank you very much!

Cheers