thebench.bio #1 Ecosystem for the Bio-Tech Industry
All Single-Cell Analysis questions

Question 15 of 20 Intermediate Troubleshooting

Cell hashing demultiplexing assigns many cells as negative or doublet. What would you check?

Short answer

Check whether the hashtag counts show one clear dominant tag per cell, whether tag concentrations were unbalanced, and whether staining or ambient tag background was the problem. Then compare the doublet rate with what the loading density predicts.

Why

Cell hashing labels each sample with an antibody or lipid tag carrying a hashtag oligo (HTO), then pools samples into one droplet run. Stoeckius and colleagues showed this lets each cell be assigned to its sample, identifies doublets formed by cells from different samples, and allows super-loading of droplet systems to cut cost. Assignment depends on each cell having one dominant tag.

The Bioconductor book's hashedDrops function reports the log fold change between the most and second most abundant tags. Confident cells have a large gap. It also adjusts for ambient tag contamination, estimated from the lower mode of each tag's distribution, because its example data showed order-of-magnitude differences in tag concentration.

Many negatives suggest weak staining, a target protein that is low on some cell types, or too little tag sequencing depth. Many doublets beyond expectation suggest a high ambient tag background, cells clumping after staining, or genuine overloading. Super-loading raises real doublets by design, but cross-sample doublets can then be removed, which was the point of the method.

How to work it through

  1. Plot the counts of each tag across cells and look for two clear modes.
  2. Plot the log fold change between the top two tags and see how many cells sit near zero.
  3. Check tag sequencing depth against the total cells captured.
  4. Compare negatives across cell types to see whether one population stains poorly.
  5. Estimate the expected doublet rate from the number of cells loaded and compare it with what was called.
  6. Confirm assignments for a subset with genotype-based demultiplexing if donors differ genetically.

An example

In a run of eight pooled donors, 30 percent of cells come out negative and 25 percent doublet. The tag plots show one tag with a weak upper mode, and almost all negatives are neutrophils, which carry less of the targeted surface protein. Genotype demultiplexing assigns 90 percent of those negatives to donors cleanly.

The team keeps the genotype calls for this run and switches to a lipid-based tag for neutrophil-rich samples.

Where the answer stops holding

Genotype-based checks only work when donors differ genetically, not for samples from one animal or one cell line. A fixed doublet expectation is approximate because clumping varies with tissue. Ambient adjustment rests on assumptions, such as a bimodal tag distribution, that fail when a tag stains poorly.

A strong answer covers

  • One dominant tag per cell is the core assumption
  • Log fold change between top two tags measures confidence
  • Tag concentration imbalance and ambient tag background distort calls
  • Poor staining of specific cell types creates negatives
  • Compare doublet calls with the expected rate for the loading

Common mistakes

  • Discarding all negatives without checking whether one cell type is affected
  • Treating every hashing doublet as a technical failure when the run was super-loaded
  • Using a single global threshold on raw tag counts despite unequal tag concentrations