Equal Weighting of Donors

Why Equal Weighting Matters

In single-cell studies, donors can contribute very different numbers of cells. That difference may reflect tissue availability, cell recovery, sequencing depth, or biological variation. If the scientific target is the average value across donors, the analysis should give each donor one contribution to that average, rather than letting donors with many observed cells dominate it.

The calculation begins after constructing one summary for each donor, such as a donor-specific mean expression level or cell-type proportion. The earlier material on donor-population estimands distinguishes the average across donors from other possible targets. Here, the focus is how to calculate that donor-average target and what assumptions make the calculation meaningful.

Definition: Let $D$ be the number of included donors, and let $\hat{\theta}_j$ be the chosen summary for donor $j$, for $j=1,\ldots,D$. The equal-weight donor estimator is $$ \hat{\theta}_{\mathrm{donor}}=\frac{1}{D}\sum_{j=1}^{D}\hat{\theta}_j. $$ Each donor summary has coefficient $1/D$, regardless of how many cells were observed for that donor.

For example, if four donors contribute 20, 50, 200, and 500 cells, each donor summary still contributes one quarter of the final estimate. The number of cells may affect how accurately a donor's summary is measured, but it does not change the donor's coefficient in this estimator. Equal weighting therefore expresses a choice of target: the average donor-level value, not a value defined by pooling all observed cells.

Target and Assumptions

Suppose donor $j$ has a donor-level quantity $\theta_j$, such as the mean expression among the relevant cells in that donor. The target for a particular set of $D$ donors is their arithmetic mean, $D^{-1}\sum_{j=1}^D\theta_j$. If the donors are regarded as sampled from a wider population, the corresponding population target is the mean of the donor-level quantity in that population. The equal-weight estimator uses $\hat{\theta}_j$, the observed summary, in place of the unobserved or population-level $\theta_j$.

Equal weighting is appropriate for that target under the following conditions. First, the donor is the biological replicate and the summary is defined consistently across donors. Second, each included donor represents one member of the target donor population, or the target is explicitly restricted to the donors observed. Third, each $\hat{\theta}_j$ estimates the intended $\theta_j$ without systematic donor-specific distortion. For example, a donor's sampled cells must represent the cell population to which that donor summary refers. Finally, the inclusion or exclusion of donors must be handled consistently with the target; silently omitting donors with difficult samples can change which donor population the estimate describes.

These conditions concern the point estimate and its target. They do not claim that all donor summaries are measured equally precisely. A summary based on many representative cells may be more stable than one based on few cells. Equal weighting deliberately does not use those differences to change each donor's contribution to the target average.

Worked Calculation With Unequal Cell Counts

Worked Example: Three Donors With Unequal Cell Counts

Suppose a hypothetical study measures a gene's normalized expression in a specified cell type. Donor A contributes 4 cells with a donor mean of 2 expression units; donor B contributes 10 cells with a mean of 4 units; donor C contributes 100 cells with a mean of 10 units. Find the equal-weight average of the three donor summaries.

The donor summaries are $\hat{\theta}_A=2$, $\hat{\theta}_B=4$, and $\hat{\theta}_C=10$ expression units. There are $D=3$ donors, so

\hat{\theta}_{\mathrm{donor}} =\frac{2+4+10}{3} =\frac{16}{3} =5.333\ \text{expression units}.

The estimate is 5.333 expression units per donor, averaged across these three donors. Donor C's 100 cells inform its own donor summary, but do not give donor C a larger coefficient in the final average. This calculation describes the mean of the three donor summaries; it does not assert that each donor has the same expression level.

The donor summaries must have a common interpretation for this average to be useful. Here, that means each is intended to summarize expression in the same cell type and on the same normalized scale. If one donor's summary instead came from a different cell type, the arithmetic would still be possible, but the result would not be the stated average of a common donor-level quantity.

Why the Formula Targets the Donor Average

A short expectation argument shows what equal weighting accomplishes. Suppose the $D$ donors are sampled from a target population and, for each donor, the observed summary is unbiased for that donor's quantity: $E(\hat{\theta}_j\mid j)=\theta_j$. Conditional on the sampled donors, linearity of expectation gives

E\!\left(\hat{\theta}_{\mathrm{donor}}\mid \theta_1,\ldots,\theta_D\right) =E\!\left(\frac{1}{D}\sum_{j=1}^{D}\hat{\theta}_j\ \middle|\ \theta_1,\ldots,\theta_D\right) =\frac{1}{D}\sum_{j=1}^{D}E(\hat{\theta}_j\mid j) =\frac{1}{D}\sum_{j=1}^{D}\theta_j.

Thus, under the stated condition, the estimator is unbiased for the average of the sampled donors' quantities. If the donors themselves are sampled representatively from a population, averaging their summaries also estimates the population's average donor-level quantity. The argument does not require the numbers of cells per donor to be equal. It does require the donor summaries to estimate their corresponding donor quantities appropriately.

The derivation also reveals a limitation: an unbiased summary for every donor is important. If cell sampling systematically over-represents a subpopulation in some donors but not others, a donor mean may be biased for the intended within-donor quantity. Giving every donor equal weight cannot repair a biased donor summary; it only specifies how the summaries are combined.

Worked Example: Equal Weighting Donor-Specific Proportions

In a hypothetical experiment, a cell type is observed in 20, 80, and 200 cells from three donors. The cell type is present in 4, 40, and 160 cells, respectively. Calculate the equal-weight average of the donor-specific proportions.

First calculate each donor's proportion: donor A has $4/20=0.20$, donor B has $40/80=0.50$, and donor C has $160/200=0.80$. Each proportion is dimensionless. The equal-weight average is

\hat{p}_{\mathrm{donor}} =\frac{0.20+0.50+0.80}{3} =\frac{1.50}{3} =0.50.

The result is an average donor-level proportion of 0.50, or 50%, for these three donors. It is not the fraction of all observed cells that belong to the cell type. The interpretation depends on the donor summaries being defined over comparable cell populations and on the observed cells being informative about each donor's proportion.

What Equal Weighting Does—and Does Not—Say

Equal weighting is not a claim that every donor contributes the same amount of information. It is a rule for defining the contribution to a particular average. The number of cells can affect the precision of an individual donor summary, while the equal-weight estimator still treats that donor as one biological replicate. A donor with very few cells may consequently have an unstable summary and the same coefficient as a donor with many cells. That is a real limitation to assess, not a reason to quietly change the target.

A different aggregation can answer a different question. For example, combining all cells before calculating a summary allows donors with more cells to contribute more to the result. In general, that is not the same as averaging the donor summaries. The choice depends on the scientific estimand: whether the intended description is an average donor or a quantity defined over the pooled observed cells. These targets should not be treated as interchangeable simply because both can be calculated from the same data.

Worked Example: Identifying a Donor Summary That Needs Caution

Consider two hypothetical donors. Donor A has 5 cells of interest, with a mean expression summary of 3 units. Donor B has 50 such cells, with a mean of 7 units. The equal-weight estimate is

\hat{\theta}_{\mathrm{donor}} =\frac{3+7}{2} =5\ \text{expression units}.

The arithmetic is correct, but the small number of cells for donor A raises a measurement concern: its mean of 3 units may be less stable than donor B's mean of 7 units. Equal weighting does not prove that the two summaries are equally precise. Before interpreting the estimate, check whether the summaries use comparable inclusion criteria, whether either donor has too few relevant cells for a meaningful summary, and whether technical or sampling problems are concentrated in particular donors. If the chosen target is the average donor, changing the weights based only on cell counts would alter the estimand; a better first step is to report the limitation and consider whether the donor summary is adequate.

Diagnostics, Failure Modes, and Alternatives

Useful checks focus on whether the donor summaries are comparable and whether the intended set of donors is represented. Tabulate the cell count and summary for every donor. Inspect donors with very low counts, missing summaries, extreme values, or visibly different cell-selection procedures. Check that the same normalization, cell-type definition, and summary rule were applied across donors. A numerical average can always be computed when values exist, but it may be scientifically misleading if donors do not represent the same quantity.

Donor-level summaries can also vary substantially. Equal weighting does not make an outlying donor disappear: each donor has the same coefficient, so an extreme summary can influence the mean. If the scientific question calls for a resistant summary, the median or another robust donor-level summary may be considered, as discussed in the earlier material on mean, median, and robust donor summaries. That choice changes the summary of donor values; it does not turn cells into independent biological replicates.

Common failure modes include:

Using cell count as a substitute for donor count. Many cells from one donor do not create additional independent donors.
Equal weighting despite incompatible summaries. Averages are hard to interpret if donor summaries refer to different cell populations, scales, or definitions.
Ignoring missing donors. An average over donors with observed summaries describes that included set; it need not represent donors excluded for low cell yield or other reasons.
Confusing a target with a precision claim. Giving donors equal coefficients defines an average donor target; it does not mean all summaries are equally precise.
Assuming equal weighting removes bias. It cannot correct systematically unrepresentative cell sampling or selection of donors into the analysis.

A practical sequence is to define the donor-level quantity, calculate one summary per eligible donor, review counts and comparability, and then average the summaries with coefficient $1/D$. Keep the target explicit in reporting. State the number of donors, describe how donor summaries were constructed, and identify important limitations such as low counts or donor exclusions. These steps make the weighting choice transparent without confusing the number of cells with the number of biological replicates.

Check Your Understanding

Answer each question using the equal-weight donor estimator and its assumptions.

Three donor summaries are 1.5, 3.0, and 7.5 expression units. What is their equal-weight average?
Donors contribute 10 and 100 cells, with donor-specific proportions of 0.2 and 0.8. What is the equal-weight average proportion?
Why does a donor contributing 500 cells not receive a larger coefficient in an equal-weight average of donor summaries?
Name one condition needed for an observed donor summary to estimate the intended donor-level quantity.
What does the equal-weight average describe if several donors were excluded because no usable summary could be calculated?

Solutions. (1) $(1.5+3.0+7.5)/3=12/3=4.0$ expression units. (2) $(0.2+0.8)/2=0.50$, or 50%; the donors each contribute one half. (3) The estimator targets the arithmetic average across biological replicates, so each donor summary has coefficient $1/D$; cell count can affect summary precision but does not redefine that target. (4) For example, the donor's observed cells should represent the cell population the summary is intended to describe, and the summary procedure should be consistent across donors. (5) It describes the average among donors with usable summaries, unless a justified sampling or missing-data approach supports a broader target; omitted donors may differ from those included.

What You'll Learn