Check complexity within the library size gradient
For clusters tracking total UMI count, inspect detected genes versus total UMIs, colored by mitochondrial fraction and split by sample. A low-complexity tail with elevated mitochondrial signal supports a quality explanation. A compact group that retains coherent complexity and appears across samples is harder to dismiss as damage. QCatch describes the expected curved relation between sequencing depth and detected genes, which makes departures from that relation informative. After plotting this, did the suspect cluster remain distinct from the low-quality tail?