Variable features of each batch of a layer-split object. — variable_features_by

After split_layers_by_batch() the per-batch variable features are stored as vf_<method>_<layer>_variable columns in the assay meta data rather than being reachable through VariableFeatures(), which returns the combined set.

Usage

variable_features_by_batch(x, assay = "RNA")

Arguments

x: Seurat object with split layers.
assay: Assay to read the variable features from.

Value

A named list of variable gene vectors, one element per batch.

Details

normalize_counts() now clears its own assay's vf_* columns before recomputing them. But an object built before that guard existed can still carry a whole history of them: each sample's own pre-merge run from create_seurat_object(), and the merge step's own post-merge preview run. DietSeurat() clears layers, reductions, and graphs, but never assay meta data. merge() disambiguates colliding names with a numeric suffix, for example vf_vst_counts.1_variable.

Matching every _variable$ column would pick up that whole history along with the batches actually being asked about. This function instead anchors on Layers(x[[assay]], search = "counts"), the object's current split state. That keeps stale columns out, regardless of which generation of the object this is.