mDAG

mDAG infers regulatory networks from mixed observational data to identify causal relationships among continuous and categorical biological variables such as gene expression, phenotypes, and single nucleotide polymorphisms (SNPs).

Key Features:

• Mixed Data Handling: Handles mixed data types, including continuous variables such as gene expression levels and categorical variables such as phenotypes and single nucleotide polymorphisms (SNPs).
• Causal Inference: Infers upstream causal factors and downstream effectors to generate hypotheses about causal directions in regulatory pathways.
• Permutation Method for Conditional Independence: Implements a permutation-based test for conditional independence tailored to mixed-type variables.
• Sparse DAG Recovery with L1 Regularization: Uses L1 regularization to recover large sparse Directed Acyclic Graphs (DAGs) from limited sample sizes.

Scientific Applications:

• Immunological Study: Inferred cytokine regulatory networks in a cross-sectional study of Chlamydia trachomatis infection.
• Cohort Study on Plasma Adiponectin Levels: Applied to a large cohort to generate mechanistic hypotheses underlying plasma adiponectin levels.

Methodology:

Uses a Directed Acyclic Graph (DAG) framework with a permutation-based conditional independence test and L1 regularization for sparse DAG recovery.