Normalised methylation data

This page outlines the steps for formatting DNA methylation data, which need to be performed before running the DEEP pipeline. This flowchart demonstrates the options you can choose to do within-cohort normalisation, and whether you are able to perform cross-cohort normalisation.

Within-cohort DNA methylation data preprocessing

Option 1 (preferred)

Ideally, we would prefer it if you normalise and QC your data using the R meffil package. Please make sure you have installed version 1.5.0 or higher. See here for installation instructions. You can check the version like this:

packageVersion("meffil")

Meffil has been optimised for speed and memory, and instructions on how to do this can be found here:

1. Perform sample QC
2. Normalise QC'd samples

Option 2

While we prefer you follow our recommended normalisation approach, we do not enforce this. If not meffil, then we would prefer that you would use an alternative functional normalisation method. If you prefer to use an alternative normalisation method then please ensure that the DNA methylation data you provide meets the following criteria:

• Beta values (i.e. proportions) should be used to quantify DNA methylation level.
• The methylation data is an R numeric matrix object where each row is a CpG site and each column is a sample.
• The rownames must be the unique cg identifiers and the column names are the IDs that correspond to the samples, and that match to sample IDs in the other datasets (genetic, covariate etc).

Data format

Regardless of which normalisation method you use, the following need to be adhered to:

• You should avoid spaces in your file name.
• You should keep sites from chrX and chrY.
• You should exclude the SNP probes with IDs that start with rsXXXX (65 on the 450K array and 59 on the EPIC array).
• You should keep nonspecific binding probes, probes with SNPs in their sequence, multimapping probes. We will filter out these probes at the end.
• Underscores should not be used in the sample IDs.
• The pipeline removes outliers with the Tukey method in the processing methylation module, so you shouldn't have NAs in your methylation dataset.
• The methylation matrix should be saved as a .RData file (each row is a CpG site and each column is a sample,) and the methylation matrix object should be called norm.beta. e.g. in R:

save(norm.beta, file="/path/to/deep/input_data/methylation.RData")

Cross-cohort DNA methylation data preprocessing

For large-scale DNA methylation studies, it is essential that data from different cohorts are processed in a consistent and comparable way. The DEEP pipeline supports this by applying a unified cross-cohort functional normalization procedure (see Module 01f Step 1). To do this, you need to have access to your idat files, and please specify the path to your idat files in your configuration file.

For the QC stages, cohorts only need to complete the within-cohort preprocessing and upload the required shrunk QC objects via the Bristol SFTP system. The resulting harmonised methylation datasets will be used in future pooled EWAS analyses through DataSHIELD in the Analysis stage.