Robust causal gene network estimation for large-scale single-cell perturbation screens using reduced control function

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Single-cell CRISPR screens are powerful for causal discovery in gene regulatory networks.
• Existing methods face limitations with confounding, count data, and high-MOI designs.

Purpose of the Study:

• Introduce RICE, a unified framework to address challenges in single-cell CRISPR perturbation screens.
• Enhance causal discovery in gene regulatory networks by overcoming limitations of current methods.

Main Methods:

• Developed a reformulated control function robust to exclusion-restriction violations.
• Integrated a constrained negative binomial model and differentiable acyclicity penalty.
• Designed a GPU-scalable model accommodating hard/soft interventions and high-MOI designs.

Main Results:

• RICE consistently outperforms existing methods in synthetic benchmarks under confounding and high-MOI conditions.
• Demonstrated improved causal discovery performance on held-out data from CRISPRi screens.
• Successfully reconstructed the interferon-gamma signaling pathway and identified novel regulatory candidates.

Conclusions:

• RICE provides a robust and scalable framework for causal discovery in single-cell perturbation genomics.
• The framework enhances the analysis of complex gene regulatory networks from CRISPR screens.
• RICE offers improved identification of regulatory relationships compared to conventional methods.