Normalize the counts, select variable features, and scale.

A thin wrapper that picks the right Seurat function for method and fixes the parameters this pipeline uses, so either path hands back a fully processed object — no separate variance step needed. method = "sct" already worked this way natively: SCTransform() produces the counts, variable features and scaled data in one pass. method = "log" now mirrors that explicitly: NormalizeData() followed by FindVariableFeatures() and ScaleData().

Usage

normalize_counts(
  x,
  method = "log",
  num_variable_genes = 3000,
  assay = "RNA",
  log_file = NULL
)

Arguments

x: A Seurat object.
method: "log" for NormalizeData() + FindVariableFeatures() + ScaleData() (log-normalized to a scale factor of 10,000), or "sct" for SCTransform().
num_variable_genes: Number of variable features.
assay: Assay to normalize.
log_file: Log file.

Value

The processed Seurat object. Also writes the variable-feature plot (variance-features.png) to the working directory.

Details

No covariates are regressed out (vars.to.regress = NULL) for either path. Germain et al. (2020) found that the "common practice of regressing out cell covariates, such as the detection rate or proportion of mitochondrial reads[,] nearly always had a negative impact."

On the log path, this function drops any vf_* columns already on assay's feature meta data before recomputing them. An earlier call can leave these columns behind: each sample's own pre-merge run, or a merge step's preview run. Left alone, they would just accumulate call after call. See variable_features_by_batch() for what reading an accumulated column list, instead of a current one, leads to.

SCTransform() never writes this kind of per-gene bookkeeping column. So the sct path has nothing to clean up. The check runs the same way for both paths; it is a no-op for sct, not something conditioned on method.