Description

This track is part of the Episignatures collection.

It shows the individual CpG sites that published DNA methylation episignatures are made of, as collected by the authors of MethaDory, an open-source tool that tests a methylation array sample against many episignature classifiers at once. The track covers more than 260,000 CpG sites drawn from 189 episignatures for over 100 rare developmental disorders, reported in 74 studies. Because a site is often part of the signature of several disorders, one row is shown per CpG site and all the episignatures that report it are listed together on that row.

All positions are Illumina methylation array probes, so the track also says something about what the arrays can and cannot see: a region with no items either carries no reported signature site or is not covered by the arrays at all.

Display Conventions and Configuration

Each item is one CpG site, named with its Illumina probe ID, and spans the two bases of the CpG dinucleotide. Typing a probe ID such as cg05654765 into the position box jumps to that site.

Items are colored by the strongest effect reported at the site. Delta-beta is the mean methylation fraction in affected individuals minus the mean in unaffected controls, so a positive value means the site is more methylated in affected individuals. Warm colors are hypermethylation, cool colors hypomethylation, and the darker shade of each pair marks a difference of at least 10 percentage points, the usual effect-size line for these arrays.

Sites shared by many episignatures nearly always contain both directions, simply because the episignatures belong to different disorders: no site reported by a single episignature disagrees with itself, but 99.7% of sites carried by eleven or more do. Flagging every such site as conflicting would color 39% of the track and would mostly be reporting how many signatures share the probe. The conflicting class is therefore restricted to sites where an opposing effect is comparable in size to the largest one.

One caution when comparing colors between disorders: the studies collected here did not use a common reporting threshold. Four published only sites with a delta-beta of at least 0.20, 25 cut at about 0.10, 33 at about 0.05, and 12 applied no visible cut, so 98% of the records come from a study that filtered at 0.05 or above. The weak classes are therefore sparse or absent for some signatures by construction rather than by biology, and a signature that looks uniformly strong may simply have been reported at a higher threshold. The Studies table below and the per-signature table on each details page show which study a value came from.

Clicking an item opens a page with a table of the episignatures at that site, one row each, giving the episignature, its disorder, its delta-beta and its adjusted p-value side by side. Rows are sorted by effect size, so the strongest episignature at the site comes first. Below the table, the PubMed entry of every study involved is linked.

The track can be filtered by disorder, by gene or locus, by study, by the direction and strength classes of the color legend, and by the largest absolute delta-beta at a site. The three list filters keep a site if any of its episignatures matches, so selecting a single disorder shows exactly the sites that belong to that disorder's signature, though the detail page of each site still lists the other disorders that share it. Zoomed out past 200 kbp the track switches to a density graph; the "Display data as a density graph" checkbox on this page overrides that.

Methods

The MethaDory authors compiled the CpG sites of every episignature published up to the date of this release, reading the site lists out of the supplementary tables of the original studies. Nearly all of these studies profiled blood-derived DNA on the Illumina Infinium HumanMethylation450 or MethylationEPIC v1.0 BeadChip, compared a small cohort of individuals carrying variants in one gene against unaffected controls, and reported the probes that came out differentially methylated together with an effect size and a p-value. Reporting practice is not standardized between these studies: some report every significant probe and some report only the subset they selected for their own classifier, effect sizes are not always computed the same way, and 81,989 of the collected rows carry no adjusted p-value, 4,335 no p-value at all. The MethaDory classifiers themselves do not use all of these sites. They apply a further quality control step that drops cross-reactive, batch-sensitive, SNP-overlapping and age-correlated probes, drops probes absent from the EPIC v2 array, and for signatures with more than 500 surviving sites keeps only the 500 most informative; support vector machines are then trained on synthetic cases generated from public control data. This track shows the collected published sites, not the reduced set the classifiers run on, and so is the more complete view of what each study reported. See Ferraro et al. for the full method.

The two source files, a table of episignature probes and a spreadsheet of signature metadata, came directly from the MethaDory authors on 16 September 2026; the tool and its models are at github.com/f-ferraro/MethaDory. The probe table gives Illumina probe IDs rather than coordinates, so positions were taken from the EPIC 850k, EPIC v2 and 450k manifests already loaded in the Genome Browser, in that order of preference. Of the 847,863 probe-by-signature records, 373 (293 distinct probes) were dropped because the probe is in none of the three manifests; the rest were merged into 268,900 CpG sites. Strand is not shown, since it is a property of the assay rather than of the CpG site. The build steps are documented in the makeDoc and the scripts are in src/hg/makeDb/scripts/episignatures.

Studies

The studies whose episignatures are included, with the number of probe records each contributed. "Probes reported" counts every probe-by-episignature record from that study and "Distinct probes" counts the CpG sites they fall on, so the two differ when a study published several episignatures that share sites.

Genes and loci

The genes and genomic regions the episignatures are derived for. Several entries are variant classes within the same gene rather than the whole gene, for instance ADNPc and ADNPt for variants inside and outside the central domain of ADNP, or NSD1 and NSD1dup for loss of function and duplication of NSD1. Entries beginning with "chr" are recurrent copy number changes.

Data Access

The data can be explored interactively in table format with the Table Browser or the Data Integrator and exported from there to spreadsheet or tab-sep tables. From scripts, the data can be accessed through our API, track=methaDory.

For bulk analysis the per-signature values are also kept as plain columns in the file, one comma-separated list each for the episignatures, disorders, genes or loci, delta-betas, adjusted p-values and studies at a site, in the same order as the rows of the table on the details page. Those columns are hidden from the details page, since the table shows the same numbers in a form that is easier to read, but they are what the Table Browser and bigBedToBed return.

For automated download and analysis, the genome annotation is stored in a bigBed file that can be downloaded from our download server. The file for this track is called methaDory.bb. Individual regions or the whole genome annotation can be obtained using our tool bigBedToBed, which can be compiled from the source code or downloaded as a precompiled binary for your system. Instructions for downloading source code and binaries can be found here. The tool can also be used to obtain features within a given range, e.g. bigBedToBed http://hgdownload.soe.ucsc.edu/gbdb/hg38/episignatures/methaDory.bb -chrom=chr21 -start=0 -end=100000000 stdout

The MethaDory tool, its trained classifiers and the underlying signature definitions are available at github.com/f-ferraro/MethaDory.

Credits

Thanks to Federico Ferraro and Dmitrijs Rots at Erasmus MC, University Medical Center Rotterdam, for compiling the episignature sites, for providing them for release under an open license, and for reviewing this track. Thanks also to the many groups whose published episignatures are collected here, listed in the table above.

References

Ferraro F, Drost M, van der Linde H, Bardina L, Smits D, de Graaf BM, Schot R, van Bever Y, Brooks A, Donker Kaat L, Brosens E, Deng R, Barakat TS, Verhoeven VJM, van Ham TJ, Kleefstra T, Rots D. Training with synthetic data provides accurate and openly-available DNA methylation classifiers for developmental disorders and congenital anomalies via MethaDory. medRxiv. 2025 Mar 30. DOI: 10.1101/2025.03.28.25324859