ComboBatch: harmonized transcriptomic matrices and benchmark metrics for 73 public germinal-centre B-cell lymphoma cohorts

Expression datasets accompanying the ComboBatch harmonization study: the public-cohort raw expression matrix, all 42 imputed input matrices (14 batch removal strategies x 3 imputation methods), and 911 harmonized matrices - including the complete 33-method x 14-strategy x 2-post-removal grid at the strict imputation - together with the full metric tables for all the 2,407 harmonization attempts.

5,178 samples are from 73 open cohorts. Fifteen further cohorts (1,996 samples) are proprietary or under controlled access and are withheld; they are named, with their sample counts, in cohorts.csv.

IMPORTANT: Every statistic in the article was computed on the full 7,174 samples - the deposited matrices are those same matrices with the withheld rows removed, not a re-harmonization of the open subset.

See README.md for the file layout, the naming convention, and the per-strategy withheld fractions. manifest.csv indexes every deposited matrix.