RSS Score Variability

I'm analyzing scRNA-seq data using pySCENIC to infer GRNs, followed by subset analysis (e.g., CD4+ T cells). I've noticed that the number of target genes per regulon can vary between subclusters biasing score direct comparisons. Should RSS be recalculated for subsets using only subset-specific target genes, or can we reuse global regulons with size correction?
