Scalable multimodal mapping of macrophage regulatory architecture by integrating optical and transcriptomic pooled screens

PerturbPair is a platform that combines parallel Perturb-seq and optical pooled screening (OPS) in primary mouse bone marrow-derived macrophages to systematically map how genetic perturbations reshape both transcriptomic and morphological cellular states at scale. We developed EB-MoCAVI, an empirical Bayesian variational inference framework that imputes RNA profiles for perturbations measured only by imaging and denoises sparse transcriptomic estimates using matched imaging data, effectively tripling the Perturb-seq compendium in silico.

Takamasa Kudo & Romain Lopez

Code organization

Folder	Contents
1-pseq-analysis/	Processing and analysis pipeline for the Perturb-seq (transcriptomic) screen: matrix formation, demultiplexing, data processing, contrastive analysis with MoCAVI, differential expression, and perturbation embedding figures.
2-ops-analysis/	Processing and analysis pipeline for the optical pooled screen (PerturbView): feature extraction, filtering, normalization, differential marker analysis, and bulk embeddings.
3-ppair-co-analysis/	Cross-modal comparison of RNA and imaging perturbation signatures, including univariate and multivariate analyses of concordance between the two modalities.
4-cross-modal-prediction-study/	Cross-modal prediction study using EB-MoCAVI, including data preparation for GEARS and scLAMBDA baselines, benchmarking, denoising, and evaluation.
5-ppair-imputed-interpretation/	Biological interpretation of measured and imputed perturbation profiles: gene signature curation, human genetics integration (UK Biobank monocyte counts, type 2 diabetes, IBD), macrophage compendium analysis, and trait correlation.
6-validation-screen/	Secondary Perturb-seq validation screen: data processing, reproducibility checks, validation of imputed profiles, benchmarking, and denoising evaluation.
library/	Shared Python library (ops and pseq modules) providing reusable utilities for differential analysis, spatial statistics, enrichment testing, gene mapping, and visualization used across all analysis folders.