getFeatureSpace() returns no features despite multiple cell types, leading to poor predictions

[Sudol67]

Hi,

I'm encountering an issue with getFeatureSpace() where no features are selected, even though the reference dataset contains many distinct cell types.

Context:

• My reference Seurat object contains ~22841 cells with 20+ well-balanced cell types
• The RNA assay is active and normalized.
• I use the standard preprocessing steps: NormalizeData, FindVariableFeatures, ScaleData, RunPCA, RunUMAP.
• The cell_type column in meta.data is correctly populated and passed to getFeatureSpace():

reference <- getFeatureSpace(reference, "cell_type")
length(reference@misc$scpred$features)  # returns 0

Consequence:
Because getFeatureSpace() does not select any features, the trainModel() step trains an incomplete model (or no model at all). As a result, scPredict() assigns only one cell type (e.g., "plasma cell") to all query cells.

Thanks in advance for any insight or workaround! I will be glad to share more information if needed