Structural factor equation models for causal network construction via directed acyclic mixed graphs.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Directed acyclic mixed graphs (DAMGs) model complex biological networks with both directed and undirected edges.
• Undirected edges in DAMGs often represent unobserved confounding factors, complicating causal inference.
• Existing methods struggle to reliably detect causal relationships obscured by latent variables.

Purpose of the Study:

• To develop an effective structural equation model (SEM) method for extracting reliable causal relationships from DAMGs.
• To introduce the structural factor equation model (SFEM) to address challenges in causal discovery within complex networks.
• To improve the interpretability of causal networks by accounting for confounding factors.

Main Methods:

• The proposed structural factor equation model (SFEM) integrates SEM for network topology with factor analysis for undirected edges.
• Latent factors are utilized within SFEM to identify and account for unobserved confounding.
• The method aims to simplify and clarify causal networks by isolating directed relationships.

Main Results:

• SFEM successfully captures network topology while accounting for undirected edges induced by latent factors.
• The method demonstrates improved identification and removal of confounding effects compared to existing approaches.
• Simulations show the effectiveness of SFEM in extracting reliable causal links.

Conclusions:

• SFEM provides a robust framework for causal discovery in the presence of unobserved confounding within DAMGs.
• The approach leads to more interpretable and accurate causal network representations.
• Application to gene regulatory networks in breast cancer highlights its practical utility.