Data dictionary and source terms • sglasso

GenAtHum

The package’s Gene Atlas Human example has 158 observations, 2,045 predictors and 79 groups. It is used in the Gaussian examples and is not the CoRSIVSZ schizophrenia dataset. Load it without a network request:

library(sglasso)

## Loading required package: Matrix

data(GenAtHum, package="sglasso")
c(observations=nrow(GenAtHum$X), predictors=ncol(GenAtHum$X),
  groups=length(unique(GenAtHum$group)))

## observations   predictors       groups 
##          158         2045           79

X is the numeric design matrix; y is the response; group maps columns to groups. gene_code, gene_name and groups_name give annotation. See help("GenAtHum") for the package’s source description.

CoRSIVSZ

CoRSIVSZ is a processed blood-methylation panel derived from public cohorts, not a newly collected cohort or a random train/test split. Original measurements are in GEO GSE84727 (development) and GSE80417 (independent external test).

Component	Meaning
`development$X`	847 by 1,107 deposited methylation values
`development$y`	414 cases and 433 controls
`external$X`	675 by 1,107 measurements in the same column order
`external$y`	353 cases and 322 controls
`group`	One integer group label per column; 409 groups
`group_name`	Original annotation cluster labels
`probe_id`	CpG identifiers matching matrix columns
`sample_id`, `geo_sample_id`	Public cohort/array and GEO sample identifiers
`preprocessing`, `provenance`	Construction rules, accessions and source records

For y, zero denotes control and one denotes schizophrenia, derived from the deposited diagnosis codes. It is not a dichotomized gene-expression outcome. Case-control proportions do not estimate population prevalence.

Groups are annotation-defined, not discovered from response performance. Each retained group has at least two original probes and every member is measured in both cohorts. Incomplete groups are excluded whole. There is no extra correlation cutoff, imputation, probe averaging, group merging or response-association filter. Deposited normalized measurements remain unchanged. Training-specific transformations belong inside model fitting/CV.

Load the separate file

The exact CoRSIVSZ_v1.rds asset is prepared locally; a public hosting endpoint has not yet been claimed. It is not inside the package tarball or the logistic run-code directory. data(CoRSIVSZ) is not supported.

# Obtain the exact versioned file from its separate distribution notice.
CoRSIVSZ <- load_CoRSIVSZ("CoRSIVSZ_v1.rds")
dim(CoRSIVSZ$development$X)
dim(CoRSIVSZ$external$X)

load_CoRSIVSZ() checks the expected size/SHA-256 before reading the file. download_CoRSIVSZ(url, destfile) requires an explicit HTTPS asset URL and refuses overwrites. Neither helper fits models or changes measurements. No data download is performed by this guide.

Original attribution

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.

The original annotation repository supplies cluster definitions. See the installed CoRSIVSZ-NOTICE.txt for source and license notices. The package’s 409-group analysis panel is not an exact reproduction of the original paper’s 1,982-region panel. Public access alone does not establish unrestricted third-party licensing or institutional ethics exemption. Retain original attribution when using these resources.