Written and maintained by CASRAI Editorial Board
Last updated
Short answer
Perturb-seq is a pooled CRISPR screening method that reads out a full single-cell RNA profile for every cell and links that profile to the guide RNA the cell received. It was introduced by Dixit et al. in Cell in 2016. One pooled experiment replaces many separate knockout-and-sequence experiments. CROP-seq, published in 2017, is a closely related variant that changes how the guide is detected.
Definition and scope
Perturb-seq combines pooled CRISPR perturbations with single-cell RNA sequencing (scRNA-seq) so that each cell’s expression profile is tied to the perturbation that cell carries. Dixit et al. demonstrated it on 200,000 cells in immune cells and cell lines, across six experiments, including bone-marrow-derived dendritic cells (BMDCs) perturbed at 24 transcription factors with and without LPS stimulation, and K562 cells perturbed at transcription factors and cell-cycle regulators.
This page covers the method as described in the two founding papers. It is background for lab staff and researchers planning or supporting a screen, not a bench protocol. For wider context see What is genomics?.
How Perturb-seq works in practice
Per Dixit et al., the workflow has three linked parts:
- Guide delivery with a barcode. A lentiviral vector delivers the sgRNA and also expresses a guide barcode (GBC), an expressed sequence that reports which sgRNA a cell received.
- Single-cell capture. scRNA-seq tags each cell’s mRNA, including the GBC transcript, with a cell barcode and a unique molecular identifier (UMI).
- Computational linking and modelling. The GBC assigns each cell to a perturbation. The authors used a linear model, MIMOSCA, to estimate per-guide, per-gene effects.
In a pilot, the method showed roughly 80% sensitivity and 90% specificity relative to bulk RNA-seq, at about 100 cells per guide.
Perturb-seq vs CROP-seq
Datlinger et al. (Nature Methods, 2017) addressed a practical problem: guide RNAs transcribed from a U6 (Pol III) promoter lack a poly-A tail, so standard poly-A-based scRNA-seq misses them. CROP-seq places the gRNA cassette in the vector’s 3′ LTR, so the gRNA becomes part of the puromycin-resistance mRNA transcribed by RNA polymerase II. The cassette is duplicated to the 5′ LTR on integration, and functional gRNAs are still expressed from the hU6 promoter.
| Question | Perturb-seq (Dixit et al. 2016) | CROP-seq (Datlinger et al. 2017) |
|---|---|---|
| How is guide identity read out? | An expressed guide barcode (GBC) linked to each sgRNA | The gRNA sequence itself, carried in a polyadenylated transcript |
| Is a separate barcode design needed? | Yes, the GBC | No; the authors state it simplifies screening with large gRNA libraries |
| scRNA-seq platform | scRNA-seq with cell barcode and UMI | Described as working with various scRNA-seq assays |
What limits a screen: planning checklist with measured numbers
The founding papers report concrete limits. These are the numbers lab staff can use as planning questions, drawn from the papers’ own experiments and not as guarantees for a new cell type.
- Can the guide be detected in this cell type? Dixit et al. detected guides in about 94% of K562 cells but about 60% of BMDCs at 3 hours, which they attribute to lower transcript complexity.
- How many cells will carry exactly one guide? In CROP-seq, unique gRNA assignment was 38.7% of cells with at least 500 detected genes, rising to 78.9% at 4,000 genes or more. The multi-gRNA rate rose from 2.7% to 9.8% across those thresholds, and multi-gRNA cells were excluded.
- How many cells per perturbation? Dixit et al. estimate about 100 to 200 cells per perturbation to detect single-gene effects, versus about 10 for signature-level effects. CROP-seq authors found about 870 cells (15%) gave results similar to the full screen.
- Is the library size realistic? Cost scales linearly with the number of perturbations, while combinations grow exponentially.
- Are filtering steps planned? Cells with multiple GBCs and chimeric reads need to be filtered. The authors also note that early pooling may cause recombination in lentivirus.
Where it matters for labs and core facilities
Because every perturbation is read in one pooled run, planning concentrates on library construction, lentiviral work, sequencing depth and analysis. Those are the points where the papers report constraints (above). Written procedures for the cell culture, transduction and library steps belong in a controlled document; see standard operating procedure (SOP) and protocols.io for how protocols are recorded and shared. Where screens produce human genomic data, GA4GH standards are relevant to data sharing.
Common misunderstandings
- “Perturb-seq and CROP-seq are unrelated.” Both link a pooled CRISPR perturbation to a single-cell transcriptome; they differ in how guide identity is recovered.
- “Every sequenced cell is usable.” Cells without a detectable guide, or with more than one, are filtered out (see numbers above).
- “More perturbations cost the same.” Cost scales linearly with the number of perturbations according to Dixit et al.
- “Pooled screens can measure anything.” Datlinger et al. note pooled screens are limited to simple readouts, while arrayed screens give transcriptomes at much lower throughput.
FAQ
What is Perturb-seq?
A pooled CRISPR screen with single-cell RNA sequencing as the readout, so each cell’s transcriptome is linked to the guide it received (Dixit et al., Cell, 2016).
What is the difference between Perturb-seq and CROP-seq?
Perturb-seq identifies the guide through an expressed guide barcode. CROP-seq places the gRNA in a polyadenylated transcript so the guide itself is read, with no barcode design.
How is Perturb-seq data analysed?
The original study used the linear model MIMOSCA to estimate per-guide, per-gene effects after assigning cells to guides through the GBC.
Sources
- Dixit A. et al. Perturb-Seq: Dissecting Molecular Circuits with Scalable Single-Cell RNA Profiling of Pooled Genetic Screens. Cell 2016. PMC5181115, read 2026-11-07.
- Datlinger P. et al. Pooled CRISPR screening with single-cell transcriptome readout. Nature Methods 2017. PMC5334791, read 2026-11-07.
