CoRSIVSZ: grouped methylation measurements for schizophrenia classification

A processed analysis panel derived from public blood DNA methylation cohorts GSE84727 and GSE80417 and the published CoRSIV/ESS/SIV cluster annotation. CoRSIVSZ names this derived panel, not a newly collected dataset or an exact reproduction of the original publication's 1,982-region panel. The full RDS file is distributed separately, not under package data/. Use load_CoRSIVSZ; data(CoRSIVSZ) is not supported.

Format

A list of class CoRSIVSZ with:

schema_version

The string CoRSIVSZ_v1.

development

A list with X (847 by 1,107 matrix), y (integer response), sample_id (public array identifiers) and geo_sample_id (GSM accessions). There are 414 cases and 433 controls.

external

The same fields for 675 independent-test participants: 353 cases and 322 controls.

group

Integer group membership for the 1,107 columns, from 1 to 409.

group_name

The 409 original annotation cluster labels.

probe_id

The 1,107 CpG identifiers, in matrix-column order.

preprocessing

An explicit description of panel construction and the unchanged deposited measurements.

provenance

Public accessions, source links and checksums.

Source

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84727

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE80417

Annotation: https://github.com/waterlandlab/CoRSIV-Methylation-based-SZ-Risk-Score

Details

The outcome is the deposited diagnosis: 0 denotes a control and 1 a schizophrenia case, corresponding to original GEO codes 1 and 2. It is not a dichotomized gene-expression outcome. The independent cohorts are not a random split, and their case-control proportions are not population prevalence estimates.

Of 2,409 annotation-defined groups, retain only groups having at least two original probes and every member measured in both cohorts. Incomplete groups are excluded whole. This produces 409 groups with sizes 2–12. No participants are removed. Columns are ordered by radix-sorted group names, then probe IDs. No response association, performance or correlation cutoff selects groups. There is no imputation, probe averaging, merging or extra smoking-probe filter.

Deposited cohort-level pfilter/dasen-normalized methylation values are unchanged. No additional scaling or covariate adjustment is applied here. Model-specific transformations and targets must be estimated only within training partitions. Public IDs establish common measurement availability, not external-outcome-driven feature selection.

Package metadata pins the exact release file size and SHA-256 digest. The complete file is approximately 11 MiB after compression and is excluded from the package tarball. Package examples never download data automatically. See the installed CoRSIVSZ-NOTICE.txt for source attribution and terms.

References

Gunasekara CJ, Hannon E, MacKay H, et al. (2021). A machine learning case-control classifier for schizophrenia based on DNA methylation in blood. Translational Psychiatry, 11, 412. doi:10.1038/s41398-021-01496-3 .

See also

load_CoRSIVSZ, download_CoRSIVSZ