Illumina methylation array probe filtering (450k and EPIC/850k)

A collection of resources to filter 'bad'/cross-reactive/variant probes from the Illumina methylation arrays during QC stages of pipelines/analysis.

450k array

BOWTIE2 mapping of 450k probes

All probe sequences were mapped to the human genome (hg19) using BOWTIE2 to identify potential hybridisation issues.

• 33,457 probes were identified as aligning greater than once
• these are made available in HumanMethylation450_15017482_v.1.1_hg19_bowtie_multimap.txt

Additional non-specific probes

Chen et al., identified a series of non-specific probes across the 450k design.

Chen Y, Lemire M, Choufani S, Butcher DT, Grafodatskaya D, Zanke BW, Gallinger S, Hudson TJ, Weksberg R: Discovery of cross-reactive probes and polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray. Epigenetics 2013, 8:203–9.

• there are a total of 29,233 probes
• these are available in 48639-non-specific-probes-Illumina450k.csv

Note: there is overlap between the two probe sets.

remember to include any probes which fail detection

# process failed probes
detP <- detectionP(RGset)
failed <- detP > 0.01
colMeans(failed) # Fraction of failed positions per sample
sum(rowMeans(failed)>0.5) # How many positions failed in >50% of samples?
failed.probes <- rownames(detP[rowMeans(failed)>0.5,])

Example filtering strategy (in R)

## generate 'bad' probes filter
# cross-reactive/non-specific
cross.react <- read.csv('48639-non-specific-probes-Illumina450k.csv', head = T, as.is = T)
cross.react.probes <- as.character(cross.react$TargetID)
# BOWTIE2 multi-mapped
multi.map <- read.csv('HumanMethylation450_15017482_v.1.1_hg19_bowtie_multimap.txt', head = F, as.is = T)
multi.map.probes <- as.character(multi.map$V1)
# determine unique probes
filter.probes <- unique(c(cross.react.probes, multi.map.probes))
## filter the matrix of beta values (beta_norm)
## CpGs probes (IlmnID) should be rownames
# fitler out 'bad' probes
table(rownames(beta_norm) %in% filter.probes)
filter.bad <- rownames(beta_norm) %in% filter.probes
beta_norm <- beta_norm[!filter.bad,]

For a real-world example filtering strategy interested parties can refer to the methods section of our publication: (http://www.genomebiology.com/2015/16/1/8)

---

EPIC/850K array

Update (200827) - added manifest revsion information

If you don't follow the Illumina website closely you may miss that the annotation manifest file goes through revision occasionally. It's important to keep an eye on this as some of these changes result in the removal of probes due to poor performance. The below table details the versions and changes. More detailed information can be found at the Illumina product page here.

Revision	Date	Description of Change
V1.0 B5	March 2020	Manifest file annotation of discordant probes
v1.0 B4	May 2017	Manifest file formatting fix
v1.0 B3	April 2017	Removed 977 CpG sites from manifest
v1.0 B2	February 2016	Fixed switch in red/green signal for Infinium I SNP probes
v1.0 B1	January 2016	Removed one pair of bisulfite conversion controls and 1031 CpG sites from the manifest - probe list
v1.0	November 2015	Initial release

Full link to the detailed change log here.

I recommend always running the latest annotation release, which is currently B5 - download.

Update (170928) - addition of probes for EPIC/850k processing

Supplementary data from Pidsley et al., (2016), suggests cross-reactive and variant containing probes to filter at QC.

Pidsley, R., Zotenko, E., Peters, T. J., Lawrence, M. G., Risbridger, G. P., Molloy, P., … Clark, S. J. (2016). Critical evaluation of the Illumina MethylationEPIC BeadChip microarray for whole-genome DNA methylation profiling. Genome Biology, 17(1), 208. https://doi.org/10.1186/s13059-016-1066-1

• there is overlap between 450k and 850k lists, however this will not cause any issues.

Extension to the above to filter EPIC data (can apply 450k list as well)

Combine the below with the above 450k process to flter EPIC arrays at QC stage:

# probes from Pidsley 2016 (EPIC)
epic.cross1 <- read.csv('EPIC/13059_2016_1066_MOESM1_ESM.csv', head = T)
# epic.cross2 <- read.csv('EPIC/13059_2016_1066_MOESM2_ESM.csv', head = T)
# epic.cross3 <- read.csv('EPIC/13059_2016_1066_MOESM3_ESM.csv', head = T)
epic.variants1 <- read.csv('EPIC/13059_2016_1066_MOESM4_ESM.csv', head = T)
epic.variants2 <- read.csv('EPIC/13059_2016_1066_MOESM5_ESM.csv', head = T)
epic.variants3 <- read.csv('EPIC/13059_2016_1066_MOESM6_ESM.csv', head = T)
# additional filter probes
epic.add.probes <- c(as.character(epic.cross1$X), as.character(epic.variants1$PROBE), as.character(epic.variants2$PROBE), 
                     as.character(epic.variants3$PROBE))
# final list of unique probes
epic.add.probes <- unique(epic.add.probes)

Filtering process follows the same as above (apply to matrix of beta values), example:

# failed probes (those that fail detection)
beta_norm <- beta_norm[!(rownames(beta_norm) %in% failed.probes),]
# additional epic probes
beta_norm <- beta_norm[!(rownames(beta_norm) %in% epic.add.probes),]