Software application profile: tpc and micd-R packages for causal discovery with incomplete cohort data.

Area of Science:

• Statistics
• Computational Biology
• Data Science

Background:

• The Peter Clark (PC) algorithm is a widely used method for data-driven causal discovery.
• Existing R implementations of the PC algorithm have limitations with missing values, temporal structures, and mixed measurement scales common in cohort data.

Purpose of the Study:

• To introduce two new R packages, micd and tpc, designed to overcome the limitations of existing PC algorithm implementations.
• To provide enhanced functionality for causal discovery using cohort data.

Main Methods:

• The micd package offers add-on functionality for handling missing values in the pcalg R package, including multiple imputation methods under the Missing At Random assumption.
• micd supports mixed measurement scales by assuming conditional Gaussianity.
• The tpc package efficiently incorporates temporal information for more informative and statistically robust causal graph outputs.

Main Results:

• The micd and tpc packages extend the capabilities of the PC algorithm for causal discovery.
• These packages enable the analysis of complex cohort data with missing values, temporal dependencies, and mixed data types.
• The developed R packages provide more accurate and reliable causal graph learning.

Conclusions:

• The micd and tpc R packages effectively address critical limitations in existing PC algorithm implementations.
• These tools facilitate more robust and accurate causal discovery from real-world cohort data.
• The packages are freely available on CRAN and GitHub, promoting wider adoption in the research community.