← Infinium annotation

MSA probe design groups

How each design group on the Infinium Methylation Screening Array (MSA) was selected: the source data, the selection criteria and how many probes carry the label.
The design fileDesignability Trait associationsCell & tissue identity Regulation & expressionAllelic, sex-linked & variable Other & unlabelled contentReferences

The design file

MSA.design.tsv.gz (Zenodo) has two columns with no header, the probe ID and its design groups, one row for each of 281,806 CpG, CpH and SNP probes. Control probes are not listed. The second column lists every group the probe belongs to, separated by commas, and colons separate a group name from its details. Some EWAS trait names contain commas themselves, so split on a comma only when the next token starts with a known group name.

The MSA design and selection criteria are described in Goldberg et al. 2025 [1]. The design file itself is a later annotation, built in 2024 by matching each probe's coordinates against the selection lists. A probe that matched none of them is labelled Other.

Counts. The "Probes" column counts probes carrying each label. A probe can carry several labels, so the column does not sum to the array size.

Designability: the shared pre-filter

All new content was drawn from one list of designable CpGs [1]. Probe sequences for both alleles were aligned to hg38 with BISCUIT. A CpG was kept only if its 30-, 35- and 40-nt subsequences and the full 50-nt probe all map with mapping quality above 20 for both alleles, which left 19,253,974 CpGs. A design score for hybridization efficiency and melting temperature had to exceed 0.3. Designs with a dbSNP 151 SNP within 5 nt of the 3′ end, or with more than six other CpGs, were removed. 9,993,793 designable CpGs remained. High-quality designs, with a score of at least 0.6, were preferred where possible.

Trait associations

GroupProbesSourceSelectionDetail format
EWAS_Hit144,6791,067 EWAS studies curated from the literature, the EWAS Catalog [2] and EWAS Atlas [3]; meQTLs from GoDMC [4] and Hawe et al. [5] High-priority studies were chosen by sample size, significance, trait diversity and citations. They contributed all their probes or their 2,500 most significant. Every other curated study contributed its top 100. Traits were grouped into 16 major trait groups by regular expression. The 10,000 CpGs with the most GoDMC meQTL associations and 20,000 random Hawe meQTLs were added. These carry the trait meqtl_hawe_na here. EWAS_Hit:trait:PMID:p-value

Cell and tissue identity

Cell-type markers were found with hierarchical one-versus-rest comparisons. Within each major cell group, a Wilcoxon rank-sum test compared the target cells with the contrast group at each CpG. Designable CpGs with the highest AUC and methylation difference were kept, balancing hyper- and hypomethylated markers where possible [1].

GroupProbesSourceSelectionDetail format
Loyfer_Tissue_Signature15,060Sorted-cell WGBS atlas, Loyfer et al. 2023 [6] Cell-type markers from the hierarchical comparison, plus the published U25 deconvolution markers from the same atlas.…:cell.in.contrast-Hyper|Hypo
BLUEPRINT_Signature12,487Sorted immune-cell WGBS [7] Immune and hematopoietic cell markers from the hierarchical comparison.…:cell.in.contrast-Hyper|Hypo
Brain_Signature16,307Single-cell brain methylomes [8–10], pseudobulked by the authors' cell-type clusters Neuronal and glial subtype markers from the hierarchical comparison. 715 entries carry the label NA.in.All-hypo, with no cell type recorded.…:cell.in.contrast-Hyper|Hypo
GeneBody_mCH_Brain_Signature2,282Genes with cell-type-specific non-CpG methylation [10] CpH probes. For each brain cell type, the ten genes with the highest AUROC were taken, and 20 CAC cytosines were sampled from each gene body.…:gene-ID_rank:gene:cell type
5hmC3,4715hmC-Seal [11] and hmC-CATCH [12] tissue 5hmC maps Tissue-specific 5hmC sites, found as for the WGBS markers. A second set spans the range of specificity: CpGs in each sample's 10,000 most-covered sites were binned by how many of 60 samples share them, with up to 200 per bin.5hmC:tissue-Hyper, or 5hmC:bin

Regulation and expression

GroupProbesSourceSelectionDetail format
ENCODE_CRE28,754ENCODE candidate cis-regulatory elements [13]; methylation range across sorted immune [7] and pan-tissue [6] cells Designable CpGs in a cCRE whose methylation varies by more than 0.4 across cell types. 30,000 were sampled, favoring enhancers: distal enhancer-like 64%, proximal enhancer-like 21%, CTCF 11%, promoter-like 2%. ENCODE_CRE:dELS|pELS|PLS|CTCF-only|…
CTCF-bound3,050ENCODE cCREs [13] CTCF-bound cCREs, chosen in the same selection as ENCODE_CRE.CTCF-bound
CpG_gene_correlation4,305Matched WGBS and RNA-seq from Roadmap Epigenomics; TCGA normal tissues Spearman correlation between each high-quality CpG and genes within 10 kb. The 2,500 most negatively and 2,500 most positively correlated CpGs were taken, ranked by significance, gene variability and expression. TCGA pairs with a correlation of at least 0.6, or at most −0.7, and P below 0.05 were added.CpG_gene_correlation:gene-ID:+|-
CpG_ATAC_correlation3,704Matched methylation and chromatin accessibility [10] CpGs whose methylation correlates with a nearby accessibility peak, with |Spearman ρ| above 0.5 and P below 0.05.CpG_ATAC_correlation:+|-
MIR200C2MIR200C locus, chr12:6,961,853–6,964,983 Designable CpGs at the MIR200C/141 promoter, an epithelial versus mesenchymal marker.MIR200C

Allelic, sex-linked and variable methylation

Groups ending in S or Sorted come from the Loyfer sorted-cell WGBS atlas [6]. The others come from bulk normal-tissue WGBS.

GroupProbesSourceSelectionDetail format
IntermediateMeth4,305180 bulk adult normal WGBS samples Autosomal CpGs with at least 20 reads and methylation between 0.3 and 0.7 in at least 140 of 180 samples. Intended to capture mono-allelic methylation.IntermediateMeth
IntermediateMethS1,528207 sorted-cell methylomes [6] CpGs between 0.3 and 0.7 in at least 180 of 207 samples, plus a curated list of imprinting CpGs.IntermediateMethS
ImprintingDMR123Skaar et al. 2012 [14], lifted to hg38 Designable CpGs in DMRs of mono-allelically expressed genes.ImprintingDMR:gene:PMID23744971
XCILinkedWGBS2,11776 female and 95 male normal WGBS samples X-linked CpGs intermediate (0.3 to 0.7) in females but unmethylated (<0.3) or methylated (>0.7) in males. These track X inactivation.XCILinkedWGBS
XCILinkedWGBSSorted2,675Sorted-cell methylomes [6] The same female-intermediate, male-binary rule, applied to sorted cells.XCILinkedWGBSSorted
Y-linked1,837180 WGBS samples; GENCODE v39 chrY genes For each chrY gene, the 20 designable CpGs most variable across samples.Y-linked:gene-IDs
CoRSIV3,316Correlated regions of systemic interindividual variation [15,16] High-quality designable CpGs inside CoRSIVs. These vary between people consistently across tissues.CoRSIV
SoloWCGW5,143Solo-WCGW CpGs in common PMDs, Zhou et al. 2018 [17] About 6,000 sampled high-quality CpGs that lose methylation with mitotic cell division.SoloWCGW

Other and unlabelled content

GroupProbesSourceSelection
Other43,742— Probes that matched no list above. They include all 4,348 SNP (rs) probes, 192 CpH probes and 39,202 CpG probes. About 14,000 of those CpGs are also on earlier human Infinium arrays.
Content without its own label. The original design also selected CpGs for several sets the 2024 annotation does not label. These include epigenetic clocks, a brain-tumor classifier, a kidney EWAS, EpiSign Mendelian-disorder episignatures, an imprintome set, transposable-element probes and Alzheimer's disease SNPs. Their probes appear under another group or under Other.

References

  1. Goldberg DC et al. Scalable screening of ternary-code DNA methylation dynamics associated with human traits. Cell Genomics 2025.
  2. EWAS Catalog, PMID 35592546.
  3. EWAS Atlas, PMID 30364969.
  4. GoDMC meQTLs, PMID 34493871.
  5. Hawe et al. meQTLs, PMID 34980917.
  6. Loyfer N et al. A DNA methylation atlas of normal human cell types. Nature 2023, PMID 36599988.
  7. Sorted immune-cell methylomes, PMID 24091925.
  8. Single-cell brain methylomes, PMID 28798132.
  9. Single-cell brain methylomes, PMID 31501549.
  10. Single-cell brain methylation, CpH and accessibility, PMID 35419551.
  11. 5hmC-Seal tissue 5hmC, PMID 33268789.
  12. hmC-CATCH tissue 5hmC, PMID 34253716.
  13. ENCODE Project Consortium. Registry of candidate cis-regulatory elements, PMID 32728249.
  14. Skaar DA et al. Imprinted DMRs, PMID 23744971.
  15. CoRSIVs, PMID 31155008.
  16. CoRSIVs, PMID 36631879.
  17. Zhou W et al. DNA methylation loss in late-replicating domains is linked to mitotic cell division. Nat Genet 2018, PMID 29610480.
Questions: wanding.zhou@pennmedicine.upenn.edu