pylimma.remove_batch_effect

pylimma.remove_batch_effect(x, batch=None, batch2=None, covariates=None, design=None, group=None, *, out_layer='batch_removed', uns_key='remove_batch_effect', layer=None, **lmfit_kwargs)[source]

Remove batch effects from a matrix of expression values.

Faithful port of R limma’s removeBatchEffect (limma/R/removeBatchEffect.R). Fits a linear model against a combined design of experimental conditions and batch covariates, then subtracts the estimated batch-coefficient contribution from the expression matrix.

Parameters:

  • x (ndarray, EList, AnnData, or dict) – Expression matrix (genes x samples) or wrapper.

  • batch (array-like, optional) – Factor of batch labels. Coded with sum-to-zero contrasts before entering the design.

  • batch2 (array-like, optional) – Second batch factor, treated the same way as batch.

  • covariates (ndarray, optional) – Quantitative covariates (samples x p). Mean-centred before entry.

  • design (ndarray, optional) – Design matrix for the experimental conditions to be preserved. If omitted and group is also omitted, a one-group design is assumed and a warning is emitted.

  • group (array-like, optional) – If given and design is omitted, sets design = one_hot(group).

  • **lmfit_kwargs – Forwarded to lm_fit (e.g. weights, method).

  • out_layer (str)

  • uns_key (str)

  • layer (str | None)

Return type:

Same class as input (matrix -> ndarray, EList -> EList, AnnData -> None).