perturb-101

A sequentially numbered, end-to-end tutorial for Perturb-seq analysis. Designed for bioinformaticians who know single-cell RNA-seq and want to learn the Perturb-seq-specific concepts, workflows, and tools.

---

Curriculum Map

#	File	Type	Topic	Dataset	Est. Time
00	00_overview.md	Markdown	What Perturb-seq is and why it exists	—	20 min
01	01_experimental_design.md	Markdown	MOI, power, controls, sequencing depth	—	25 min
02	02_guide_design.md	Markdown	gRNA scoring, TSS targeting, barcodes	—	20 min
03	03_data_acquisition.ipynb	Notebook	Download and inspect public datasets	Norman 2019, Replogle 2022	1.5 hr
04	04_raw_processing_overview.md	Markdown	Cell Ranger / STARsolo / kb-python	—	30 min
05	05_quality_control.ipynb	Notebook	scRNA-seq QC + guide UMI QC	Norman 2019	2 hr
06	06_guide_assignment.ipynb	Notebook	Assign guides to cells; multiplet handling	Norman 2019	2 hr
07	07_normalization_dimred.ipynb	Notebook	NTC-anchored PCA, cell cycle, UMAP	Norman 2019	2 hr
08	08_perturbation_effects.ipynb	Notebook	Pseudobulk DE (PyDESeq2), effect sizes	Replogle 2022	3 hr
09	09_mixscape.ipynb	Notebook	Escape detection with Mixscape	Norman 2019	2 hr
10	10_visualization.ipynb	Notebook	Volcano, heatmap, perturbation UMAP	Both	2.5 hr
11	11_genetic_interactions.ipynb	Notebook	Additive model, interaction scores	Norman 2019	3 hr
12	12_grn_and_prediction.ipynb	Notebook	TF activity, GEARS, linear baseline	Both	3–5 hr
—	glossary.md	Markdown	Key terms A–Z	—	Reference

Total estimated time: ~25–30 hours

---

Prerequisites

• Python 3.11+, conda
• Familiarity with scRNA-seq concepts (UMIs, barcodes, normalization, UMAP)
• Experience with scanpy and anndata is helpful but not required

---

Setup

# Clone and enter the repo
git clone <repo-url>
cd perturb-101

# Create the conda environment
conda env create -f environment.yml
conda activate perturb-101

# Launch JupyterLab
jupyter lab

---

Data

All data is downloaded programmatically in notebook 03. Two public datasets are used:

Dataset	Paper	Source	Size
Norman et al. 2019	Science 365, eaax4438	Figshare	~109k cells, 236 perturbations
Replogle et al. 2022 (K562 Essential)	Cell 185, 2117–2128	Figshare	~650k cells, 2,057 perturbations

Downloaded files are stored in data/ (gitignored). Notebook 03 also saves a smaller Norman subset for fast iteration.

---

Notebook Dependencies

03 (download)
  └── 05 (QC)
        └── 06 (guide assignment)
              └── 07 (normalization)
                    ├── 08 (perturbation effects) ← uses Replogle
                    └── 09 (Mixscape)
                          └── 10 (visualization)  ← uses both 08 + 09 outputs
                                └── 11 (genetic interactions)
                                      └── 12 (GRN + prediction)

Notebooks 08 and 09 are independent and can be run in parallel after 07.

---

Key Packages

Package	Purpose
scanpy	Core scRNA-seq analysis
anndata	Data container
pertpy	Guide assignment, Mixscape, distances, Augur
pydeseq2	Pseudobulk differential expression
decoupler-py	TF activity inference
pooch	Reproducible file downloads
muon	Multi-modal AnnData (MuData)

---

Citations

If you use these materials, please also cite the datasets:

• Norman, T.M. et al. Exploring genetic interaction manifolds constructed from rich single-cell phenotypes. Science 365, eaax4438 (2019).
• Replogle, J.M. et al. Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. Cell 185, 2117–2128 (2022).
• Heumos, L. et al. pertpy: an end-to-end framework for perturbation analysis. Nature Methods (2025).