Genome-Scale Perturb-seq Re-analysis

Re-analysis of the K562 CRISPRi genome-wide Perturb-seq dataset from Replogle et al., Cell 2022.

Data

Uses pre-computed Z-normalized pseudo-bulk profiles from Figshare. The notebook will download it automatically on first run (~370 MB).

What's in the notebook

• QC of guide representation and batch effects
• Z-score validation and multi-guide concordance
• PCA-based state shift quantification (Euclidean + Mahalanobis)
• Hit scoring with permutation testing and FDR correction
• Pathway analysis (decoupler ULM), NMF gene programs, GSEA
• Binary and multi-class classification for predictive validation
• Sensitivity analysis: cell dropout, weighting schemes, leave-one-batch-out
• Screen design considerations for in vivo translation
• Translational priority scoring based on pathway relevance

Requirements

scanpy anndata numpy pandas scipy scikit-learn matplotlib seaborn
xgboost statsmodels adjustText decoupler gseapy pertpy

Usage

Open replogle2022_perturbseq_analysis.ipynb in Jupyter and run all cells. Data downloads automatically on first run.