Applying causal discovery to single-cell analyses using CausalCell.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Correlation does not imply causation, necessitating causal discovery methods for biological insights.
• Single-cell RNA-sequencing (scRNA-seq) data offers a powerful resource for inferring causal relationships.
• Challenges exist in selecting and applying causal discovery methods to complex single-cell data.

Purpose of the Study:

• To develop and present a standardized workflow and platform, CausalCell, for single-cell causal discovery.
• To benchmark various causal discovery algorithms for their efficacy on scRNA-seq datasets.
• To provide guidance and best practices for applying causal inference in single-cell studies.

Main Methods:

• Benchmarking of four distinct causal discovery methods using simulated and real scRNA-seq data.
• Development of the CausalCell platform to facilitate reproducible single-cell causal inference.
• Application of the workflow to multiple scRNA-seq datasets to evaluate method performance.

Main Results:

• No single causal discovery method is universally optimal; method choice depends on the specific biological context.
• The constraint-based PC algorithm, utilizing kernel-based conditional independence tests, demonstrated superior performance across most tested scenarios.
• The study provides empirical evidence supporting the utility of causal discovery in unraveling gene regulatory networks from scRNA-seq data.

Conclusions:

• CausalCell offers a robust framework for performing single-cell causal discovery.
• The PC algorithm with kernel-based tests is a recommended approach for many single-cell causal inference tasks.
• Inferred causal interactions can significantly advance the understanding of molecular mechanisms, gene regulation, and therapeutic target identification in single cells.