MRPC: An R Package for Inference of Causal Graphs.

Area of Science:

• Computational Biology
• Genomics
• Causal Inference

Background:

• Causal relationships are key in scientific inquiry, often modeled as directed graphs.
• Existing graph inference methods yield many false positives, and genomic causal inference methods are limited.
• Gene regulatory networks are an example of causal networks in biology.

Purpose of the Study:

• To present MRPC, an R package for accurate causal graph learning.
• To improve upon existing causal inference methods, especially for genomic data.
• To provide a robust tool for general and biomedical causal discovery.

Main Methods:

• MRPC builds on the PC algorithm for learning directed acyclic graphs.
• It incorporates Mendelian randomization principles as constraints for edge orientation in genomic data.
• The algorithm enhances accuracy in identifying v-structures (X→Y←Z) and is robust to data arrangement.

Main Results:

• MRPC demonstrates improved accuracy in causal graph learning compared to existing methods.
• It offers increased accuracy in identifying v-structures.
• The package provides robustness to node arrangement in input data.

Conclusions:

• MRPC is a powerful R package for inferring causal graphs with enhanced accuracy.
• It is particularly effective for biomedical and genomic data by leveraging Mendelian randomization.
• The open-source MRPC package is available on CRAN for general use.