Sample deliverable — the reviewer response

Two comments come back on almost every single-cell manuscript. Neither is answered by argument; both are answered by running something and reporting the number. Here they are, worked through on public data (public 10x Genomics PBMC 3k) exactly as they would go into a rebuttal.

Comment 1 — “The QC thresholds appear arbitrary.”

The reviewer is not asking for different thresholds. They are asking whether your conclusions depend on them. So we re-run the whole pipeline at two alternative threshold sets and report what changes.

Run	Thresholds	Cells	Clusters	ARI vs published	Canonical markers recovered
as published	genes 200–2500, mito <5%	2,638	6	1	6
stricter	genes 300–2000, mito <3%	2,246	6	0.894	6
more permissive	genes 150–4000, mito <15%	2,698	7	0.872	7

What this particular run actually showed

Cluster structure is stable (ARI 0.894 and 0.872 against the published thresholds) — but the permissive run recovers one extra cluster and with it PPBP, the platelet marker, which the stricter filter removes entirely. The strict run loses a real, small population.

That is the honest answer to the comment, and it is more useful than "our thresholds are standard": the main conclusions do not move, and one rare population is threshold-dependent, so it is reported with that caveat rather than as a clean finding.

How it goes into the rebuttal

“We repeated the entire analysis at two additional quality-control settings (stricter: 300–2,000 genes per cell, <3% mitochondrial; more permissive: 150–4,000 genes, <15% mitochondrial). Cluster structure was stable across settings (adjusted Rand index 0.894 and 0.872 relative to the reported analysis) and all major populations were recovered in every run. The permissive setting additionally retained a small platelet population (PPBP+), which is excluded by stricter mitochondrial filtering; we have noted this dependency in the revised Results. A sensitivity table is provided as Supplementary Table S1.”

Comment 2 — “Were doublets identified and removed?”

A gene-count ceiling is not doublet handling. We run a detector, report the expected and observed rates, and show where the doublets sit.

2.2%expected at this cell count

1.18%detected (31 of 2,638)

8.33%in the most affected cluster

Cluster	Predicted doublets
Cluster 0	0.17%
Cluster 1	3.82%
Cluster 2	1.88%
Cluster 3	0.24%
Cluster 4	8.33%
Cluster 5	0%

Detected rate sits below the rate expected from loading, and the doublets are not spread evenly — cluster 4 carries 8.33% of them. A cluster that concentrated doublets at several times the background rate would be a cluster to re-examine before calling it a novel state, and that is exactly the check the reviewer is asking for.

Why this service exists

Revision deadlines are short, the original analyst has often moved on, and the comments usually require re-running work rather than rewriting text. We reproduce the existing analysis, run what is being asked for, and hand back both the figures and the paragraphs — with the code, so the next round costs nothing extra. Indicative price for this tier is $450–1800, fixed in writing before we start.

→ Send us the comments · the five requests reviewers actually make · the script that produced this page