Specimen-Duplication Detection Across Cohorts — os_detect_cross_cohort

Detects putative specimen duplicates across multiple cohorts/datasets by either (a) exact feature equality on a chosen fingerprint matrix, or (b) an explicit crosswalk of internal specimen IDs.

In circulating-miRNA cancer meta-analyses this is critical because the same biological samples are often submitted to GEO under different series accessions with different GSM IDs. Failing to dedup leads to specimen-level train/test leakage and anti-conservative bootstrap CIs (documented in GSE106817/GSE113486/GSE113740 for ovarian cases).

Usage

os_detect_cross_cohort_duplicates(
  X,
  cohort,
  sample_id,
  specimen_id = NULL,
  tolerance = 0,
  min_features = NULL
)

Arguments

X: Numeric matrix with one row per sample, columns are features.
cohort: Vector of cohort identity per sample.
sample_id: Vector of sample IDs.
specimen_id: Optional explicit internal specimen IDs (preferred).
tolerance: Numeric tolerance for exact equality on feature vectors. Default 0 (numeric equality without rounding).
min_features: Minimum number of jointly finite features required to declare a match. All jointly finite features must agree within `tolerance`. Default: all columns, which therefore requires complete profiles.

Value

A data frame with columns: sample_id_a, cohort_a, sample_id_b, cohort_b, specimen_id (if provided), match_type.