BatchVaria: variance-based evaluation of batch correction with provenance tracking

Assessing batch correction requires reproducible comparison of correction configurations and their effects on technical and biological variation. We present BatchVaria, an R package that integrates correction, variance profiling and reassessment within SummarizedExperiment objects. It retains original and corrected assays, records analysis provenance, variance denominators and summarisation procedures, and provides extensible interfaces for correction methods and variance engines. An application to TCGA breast cancer RNA-seq data demonstrates these capabilities across four correction configurations. Subtype variance fractions increase after every correction. Combining pooled fractions with recorded assay totals reveals that the absolute subtype contribution decreases by 4.23% under naive ComBat, whereas it is essentially unchanged (+0.06%) under subtype-protected ComBat. Under an additive plate-plus-subtype model, both limma::removeBatchEffect configurations have the same adjusted subtype contribution as the uncorrected assay. Their signed shared remainders nevertheless differ (-0.62% without subtype protection; +0.66% with protection), revealing a difference in attribution that the adjusted contribution alone does not capture. BatchVaria thereby supports comparison of correction configurations using absolute contributions and attribution diagnostics, rather than variance fractions alone.