The novel cell type that was a doublet

The problem

The lab had submitted a manuscript describing a new tumor-infiltrating population that co-expressed CD3E, CD8A, CD68 and C1QB. The cells sat between the T-cell and myeloid clusters on the UMAP and were reported as evidence of fusion or phagocytic transfer. Reviewer 2 asked for doublet analysis and ambient RNA correction. The original analyst had left the lab, and the processing code was incomplete.

What we found

• We re-ran alignment from FASTQ and found the original filtering had used a single global threshold that kept many high-UMI barcodes.
• scDblFinder and DoubletFinder, run per sample, flagged 71% of the 'hybrid' cells as doublets. Their UMI counts were about 1.9 times the median of either parent population, which is the signature of two cells in one droplet.
• CellBender showed strong ambient contamination with myeloid transcripts (C1QA/B/C, LYZ, APOE) in two patients with heavily necrotic samples. That explained most of the remaining 'hybrid' signal.
• After correction, the population dissolved into its parents, except for a small set of cells with a coherent, reproducible program.

How we rescued it

1. Rebuilt the pipeline end to end in Nextflow, with per-sample adaptive QC, ambient correction and doublet removal.
2. Re-integrated with scVI and benchmarked integration against Harmony using scIB metrics.
3. Re-annotated all compartments with a cell biologist. The small, coherent set turned out to be a TOX-high, CXCL13-positive exhausted CD8 T-cell state that expanded in non-responders.
4. Replaced cell-level Wilcoxon tests with pseudobulk DESeq2 across patients, and Milo for differential abundance.
5. Delivered a point-by-point reviewer response, revised figures and a reproducible methods section.

The result

The authors withdrew the hybrid-cell claim and reframed the paper around the exhausted state. That finding held up across patients and in a public validation cohort. The manuscript was accepted after one more round of minor revisions.