How Many Cells for Single-Cell Sequencing? Rare Cell Type Calculator

A cell type makes up a fraction p of your sample and you want at least k of them with 95% confidence — how many cells must pass QC? Exact binomial calculation, plus the reverse: your chance of success for a given number of cells.

Everything runs locally in your browser. Pasted data is never uploaded to any server.

1. Goal

Fraction of this cell type (%)Minimum count, k

Confidence (%)

2. Cells to load (optional)

Share of loaded cells that survive to post-QC (%)

Losses from capture efficiency, doublets and QC filtering depend on your platform, sample and protocol, so no number is built in here. Use your platform’s documentation and your own past runs. The default of 100% means no adjustment.

3. Reverse calculation (optional)

Post-QC cell count N

If filled in: the probability of capturing at least k cells of the type among this many cells.

Result

Download

Model: each cell is treated as an independent random draw from the sample, so the count of a type is X ~ Binomial(N, p). The answer is the smallest N with P(X ≥ k) ≥ the confidence level — the same as k + qnbinom(c, size = k, prob = p) in R, so you can check it yourself. The sums are exact; no normal or Poisson approximation is used.
Why not just k ÷ p? That only gives k cells on average; the actual count varies, so planning on the average gives a clearly lower chance of success than the confidence you want (the result shows the exact value for your inputs).
Assumptions: random sampling; the fraction is known, the same in every sample and unaffected by treatment; no doublets or batch effects, and no cell-type-specific losses during dissociation and QC (neutrophils and neurons, for example, are lost more easily). Real experiments are usually worse, so read the result as a minimum.
If the fraction itself is uncertain (from literature, flow cytometry or a pilot), use the low end of the plausible range (a conservative value). The table under the result shows how many cells you need if the true fraction is lower.
“Cells to load” is just your post-QC number divided by the retention you enter. Capture efficiency, doublet rate and QC pass rate vary with platform, sample and protocol, so this page offers no defaults; rely on your platform’s documentation and your own data.
With several cell types, the largest of the per-type requirements is used; the chance that all succeed together is lower — possibly much lower with many types or little margin — and the page shows a conservative lower bound for it.
This only answers how many cells you capture. It says nothing about sequencing depth per cell, and it is not a sample-size or power calculation for differential analysis.