Inferring Regulatory Networks From Mixed Observational Data Using Directed Acyclic Graphs

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Inferring gene regulatory networks from gene expression data is crucial but challenging.
• Existing Directed Acyclic Graph (DAG) methods often fail to model mixed continuous and categorical data.
• This limitation restricts joint analysis of gene expression and other variable types.

Purpose of the Study:

• To introduce a new mixed DAG (mDAG) algorithm for inferring regulatory pathways from mixed observational data.
• To enable joint modeling of continuous gene expression and categorical variables (phenotypes, SNPs).
• To identify upstream causal factors and downstream effectors in regulatory pathways.

Main Methods:

• Developed a novel mDAG algorithm for mixed-type data.
• Introduced a permutation method for testing conditional independence of mixed variables.
• Applied L1 regularization for recovering sparse DAGs with limited sample sizes.
• Validated performance through extensive simulations against existing methods.

Main Results:

• mDAG significantly outperformed two established methods in recovering the true underlying DAG in simulations.
• Successfully inferred the cytokine regulatory network in a cross-sectional study of Chlamydia trachomatis infection.
• Generated mechanistic hypotheses for plasma adiponectin levels in a large cohort study.

Conclusions:

• The mDAG algorithm provides a robust framework for constructing regulatory networks from mixed data.
• It offers a powerful tool for hypothesis generation in complex biological systems.
• The R package mDAG is available for public use.