scCausalVI

scCausalVI is a causality-aware generative model for analyzing perturbational single-cell RNA sequencing data. The model addresses a fundamental challenge in single-cell analysis: disentangling intrinsic cellular heterogeneity from treatment-induced effects, in particular for case-control study. By incorporating structural causal models with deep learning, scCausalVI:

• Learns disentangled and interpretable latent representations that separate baseline cellular states from treatment effects

• Models cell-state-specific responses to explore differential response pattern

• Enables in silico perturbation to predict cellular states under alternative experimental conditions

• Integrates multi-source data while distinguishing batch effects from biological signals (baseline states and treatment effects)

• Identifies treatment-responsive populations and characterizes molecular signatures of susceptibility and resistance to disease

The framework supports comprehensive downstream analyses, including clustering, visualization, differential expression analyses, and cross-condition prediction, providing researchers with tools to investigate cellular heterogeneity and treatment responses at single-cell resolution.

Contents:

• scCausalVI
• Installation
• Tutorial
  • Data Pre-processing
  • Visualize latent data
  • Prediction accuracy
  • Cross-condition prediction
  • Permutation test to identify responsive cells in stimulated cohort
  • Analyze differential genes between susceptible and resistant cells
  • GO term analysis with differentially expressed genes

Website and documentation: https://sccausalvi.readthedocs.io/.

Source Code (MIT): https://github.com/ShaokunAn/scCausalVI/.

A demo data is available at source data link.

Author’s Homepage: https://shaokunan.github.io.

Installation

The scCausalVI package can be installed via pip:

pip install scCausalVI