Cell Hashing

By labeling cells from different samples with distinct hashtag oligonucleotides (HTOs) , all samples can be pooled into a single droplet-based run. This eliminates technical variation between samples, reduces reagent costs, and dramatically increases throughput by processing multiple conditions simultaneously.

After sequencing, cells are assigned back to their sample of origin based on HTO read counts. Algorithms fit negative binomial distributions or use k-medoids clustering on HTO count matrices to classify cells. This replaces error-prone physical sorting with precise, probabilistic assignment.

Cells with high counts for two or more distinct HTOs are flagged as doublets or multiplets—two cells encapsulated in one droplet. This provides a direct, antibody-based measurement of cross-sample multiplets that is orthogonal to transcriptome-based doublet detection, enabling high-confidence removal.

The core reagent is a monoclonal antibody targeting a ubiquitously expressed surface protein (e.g., CD45, CD298, or β2-microglobulin), covalently attached to a DNA barcode flanked by PCR handles. This ensures every cell in a sample receives the same identifying tag regardless of cell type.

Cell hashing can be combined with CITE-seq in a single workflow. Hashtag antibodies label sample origin, while a separate panel of oligonucleotide-conjugated antibodies quantifies surface protein expression. This yields multimodal data—transcriptome, protein, and sample identity—from each cell.

Pooling samples reduces per-cell library preparation and sequencing costs. Instead of running 8 separate 10x Genomics channels, a single channel processes all 8 hashed samples. This makes large-scale perturbation screens, patient cohort studies, and time-course experiments economically feasible.

The algorithmic identification of erroneous profiles from two or more cells encapsulated in a single droplet. Cell hashing provides a direct experimental method for doublet detection: droplets containing cells with different hashtag oligonucleotides are flagged as multiplets. Computational methods like Scrublet and DoubletFinder offer complementary approaches when hashing is not used.

The computational alignment of multiple single-cell datasets into a shared latent space. While cell hashing eliminates batch effects by processing samples together physically, data integration methods like Harmony and scVI correct for technical variation when samples must be processed separately. Hashing and integration are often used together in large-scale atlas projects.

A computational process that removes technical variation introduced by different experimental batches. Cell hashing minimizes batch effects at the source by enabling all samples to be processed in a single tube. When hashing is not feasible, methods like ComBat and MNN Correct adjust expression values post hoc to align distributions across separately processed samples.

A supervised approach that projects cell-type annotations from a reference atlas onto a query dataset. After demultiplexing hashed samples and clustering cells, label transfer accelerates annotation by mapping clusters to known identities. Tools like Seurat's FindTransferAnchors and SingleR automate this process, reducing manual marker gene curation.

A preprocessing step that adjusts raw gene expression counts for sequencing depth and capture efficiency. After demultiplexing hashed samples, each sample's count matrix must be normalized before comparison. Methods include library-size normalization, SCTransform (regularized negative binomial regression), and scran pooling-based size factor estimation.