ComBat batch correction

Adjusts batch effects with the empirical-Bayes ComBat method (Johnson et al. 2007), applied to one of "intensity", "norm_intensity", or "conc". ComBat models a location and scale batch effect per feature and shrinks those estimates across features, which can stabilise many small batches better than simple median centering.

Unlike correct_batch_centering() and correct_batch_serrf(), ComBat estimates batch effects from all samples (optionally protecting biology via covariates), not from the reference QCs. On strongly unbalanced designs this can remove genuine biological signal, so supply covariates when the biological grouping is not balanced across batches. ref_qc_types is used only for the before/after QC-CV report and the plotting trend curves.

Batch correction is performed after normalization and drift correction in the recommended pipeline. Features with any missing or non-finite values in the selected variable are left uncorrected (ComBat requires complete data).

Usage

correct_batch_combat(
  data = NULL,
  variable,
  ref_qc_types,
  covariates = NULL,
  ref_batch = NULL,
  parametric = TRUE,
  replace_previous = TRUE,
  log_transform_internal = TRUE,
  feature_list = NULL,
  replace_exisiting_trendcurves = FALSE
)

Arguments

data

A MRMhubExperiment object.

variable

The variable to correct: one of "intensity", "norm_intensity", or "conc".

ref_qc_types

Character vector of QC types used for the QC-CV report and trend curves (not for the ComBat fit itself).

covariates

Optional model matrix of biological covariates to preserve (passed to sva::ComBat() as mod). Defaults to NULL (no covariates).

ref_batch

Optional reference batch to adjust the others towards (passed to sva::ComBat() as ref.batch). Defaults to NULL.

parametric

Use the parametric empirical-Bayes prior (TRUE, default) or the non-parametric prior (FALSE).

replace_previous

Replace a previous batch correction (TRUE, default) or apply on top of it.

log_transform_internal

Fit ComBat in log10 space (TRUE, default, appropriate for multiplicatively-scaling MS data). Returned data are always back-transformed to the raw scale.

feature_list

Optional feature selection (character vector or a single regular expression); NULL (default) selects all features.

replace_exisiting_trendcurves

Reseed the plotting trend curves. Default FALSE.

Value

A MRMhubExperiment with corrected data.

References

Johnson WE, Li C, Rabinovic A (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics, 8(1), 118-127. doi:10.1093/biostatistics/kxj037

Applied through sva::ComBat() from the sva package. See also Broadhurst D, et al. (2018), Metabolomics, 14, 72 (doi:10.1007/s11306-018-1367-3 ) on QC-based signal correction.

See also

correct_batch_centering(), correct_batch_serrf(), correct_drift_loess() and plot_runscatter() for visualisation. The drift and batch correction manual.

Examples

# mexp <- correct_batch_combat(mexp, variable = "conc", ref_qc_types = "BQC")