Causal Inference for Genomic Data with Multiple Heterogeneous Outcomes

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is a standard genomics approach.
• Causal inferences at the cohort level are now possible with scRNA-seq.
• Gene expression levels are not directly observable, requiring estimation from proxy measurements.

Purpose of the Study:

• Propose a generic semiparametric inference framework for doubly robust estimation.
• Address causal inference with multiple derived outcomes in genomics.
• Quantify causal effects of heterogeneous outcomes using standardized average treatment effects and quantile treatment effects.

Main Methods:

• Developed a semiparametric inference framework for doubly robust estimation.
• Specialized analysis for standardized average treatment effects and quantile treatment effects.
• Utilized Von Mises expansions and estimating equations for estimators.
• Implemented a Gaussian multiplier bootstrap for multiple testing to control false discovery exceedance rate.

Main Results:

• Demonstrated the utility of semiparametric inferential results for doubly robust estimators.
• Showcased applications in single-cell CRISPR perturbation analysis.
• Provided insights into using different estimands for causal inference in genomics.

Conclusions:

• The proposed framework offers a robust method for causal inference in genomics using scRNA-seq data.
• The methods are applicable to various genomic analyses, including perturbation studies and differential expression.
• The study highlights the importance of appropriate estimands for reliable causal effect quantification.