Cell Hashing

Cell hashing relies on antibodies conjugated to oligonucleotide barcodes rather than fluorophores. These antibodies target ubiquitously expressed surface proteins—typically CD45 for immune cells or CD298 (β2-microglobulin) for broader cell types. Each sample receives a distinct barcode sequence, allowing up to 12 or more samples to be pooled into a single single-cell sequencing run. The oligo tag contains a PCR handle for amplification, a unique barcode for sample identification, and a poly-A tail for capture on droplet-based platforms.

After labeling, cells from different samples are pooled into a single tube for downstream processing. This multiplexing strategy dramatically reduces technical batch effects because all samples experience identical reagent exposure, droplet encapsulation, and sequencing conditions. Key advantages include:

• Cost reduction: Up to 8-12x fewer sequencing lanes required
• Throughput increase: Process dozens of samples simultaneously
• Batch effect elimination: No need for post-hoc data integration
• Doublet detection enhancement: Cross-sample doublets are immediately identifiable

After sequencing, cells are assigned back to their sample of origin using HTO (Hashtag Oligo) counts. The demultiplexing algorithm evaluates the relative abundance of each barcode per cell barcode. A cell is classified as a singlet when one HTO signal dominates, a doublet when two HTOs show strong signal, and negative when no HTO is detected. Tools like Seurat's HTODemux and CiteFuse use Gaussian mixture models or k-medoids clustering on HTO count distributions to make these assignments with high confidence.

A critical advantage of cell hashing is its ability to identify cross-sample doublets—droplets containing cells from two different samples. These are trivially detected because they exhibit strong signal for two distinct HTO barcodes. This complements traditional doublet detection methods that rely on gene expression features alone. The combination provides:

• Higher sensitivity for heterotypic doublets
• Ground truth validation for computational doublet callers like Scrublet or DoubletFinder
• Cleaner cell-type clusters after doublet removal
• Reduced false discoveries in differential expression testing

Cell hashing is frequently combined with CITE-seq or TotalSeq workflows, where cells are simultaneously labeled with HTO antibodies and ADT (Antibody-Derived Tag) antibodies targeting specific surface proteins. This enables:

• Sample multiplexing via HTOs
• Surface protein quantification via ADTs
• Transcriptome profiling via mRNA capture All three modalities are captured in a single sequencing run, providing a rich multimodal dataset. The Seurat v5 framework natively supports joint analysis of RNA, HTO, and ADT assays through its multi-assay data structure.

Cell hashing introduces specific QC metrics beyond standard scRNA-seq parameters. Analysts must evaluate:

• HTO signal-to-noise ratio: The enrichment of HTO counts over ambient background
• Singlet purity: The confidence score of sample assignment
• Cross-contamination rate: The fraction of cells with ambiguous HTO profiles
• Negative cell percentage: Cells with insufficient HTO signal, often dead or damaged Cells failing these metrics are filtered before downstream analysis. Ambient HTO contamination, analogous to ambient RNA, can be corrected using algorithms like HTODemux's negative binomial model or SoupX adapted for HTO counts.

The core reagent in cell hashing. These are monoclonal antibodies conjugated to a short, sample-specific DNA barcode rather than a fluorophore. They target ubiquitously expressed surface proteins (e.g., CD45 for immune cells, β2-microglobulin). During library preparation, the barcode is amplified and sequenced alongside the transcriptome, labeling each cell's sample of origin.

Cells from distinct biological samples are incubated with their respective hashtag antibodies, washed to remove unbound reagent, and then combined into a single pool before droplet encapsulation. This physical mixing eliminates the need for separate library preparations and sequencing runs, reducing technical batch effects and reagent costs proportionally to the number of samples multiplexed.

The algorithmic process of assigning each sequenced cell barcode back to its original sample using hashtag oligonucleotide counts. Key steps include:

• Count Matrix Construction: Building a hashtag-by-cell count matrix from aligned reads.
• Signal Normalization: Centered log-ratio (CLR) normalization across hashtags per cell.
• Thresholding: Classifying cells as 'singlets' (one dominant hashtag), 'doublets' (two), or 'negative' (unlabeled debris). Tools like Seurat's HTODemux and CiteFuse automate this workflow.

A powerful quality control benefit unique to multiplexing. True biological doublets formed from cells of different samples will express two distinct hashtag oligos. This provides a ground-truth training set for doublet detection algorithms that is impossible to obtain from transcriptomic data alone. These verified cross-sample doublets are used to calibrate and validate computational doublet finders like Scrublet or DoubletFinder.

The specific term for the antibody-derived tag (ADT) used in cell hashing. The HTO count matrix is processed in parallel with the RNA count matrix. A critical quality metric is the HTO signal-to-noise ratio, defined as the enrichment of the dominant hashtag over background. Cells with low signal are flagged as unassigned, ensuring only confidently labeled singlets proceed to downstream differential expression analysis.

The practical limit of how many samples can be pooled. Efficiency depends on:

• Hashtag Diversity: The number of orthogonal, non-cross-reacting barcodes available.
• Sequencing Saturation: Allocating sufficient reads to resolve all hashtags.
• Doublet Rate: The probability of forming cross-sample doublets increases linearly with the number of samples loaded, requiring a balance between throughput and data loss.