Dynamical Causality Under Latent Confounders for Biological Network Reconstruction.

Area of Science:

• Systems Biology
• Computational Biology
• Network Science

Background:

• Causal interaction inference is challenged by spurious interactions due to biological system confounders.
• Existing methods struggle with inferring causal interactions under latent (unobserved) confounders.

Purpose of the Study:

• To develop a method for inferring dynamical causality under latent confounders.
• To reconstruct latent confounders from time-series data.
• To address the long-standing problem of causal detection in high-dimensional systems with limited observed variables.

Main Methods:

• Proposed a novel method based on an orthogonal decomposition theorem in a delay embedding space.
• Developed a theoretical foundation for causal detection in high-dimensional systems.
• Enabled separation of coupled variables in the embedding space to solve non-separability issues.

Main Results:

• Successfully inferred dynamical causality under latent confounders.
• Effectively reconstructed latent confounders from time-series data.
• Demonstrated the method's capability for causal detection even with only two observed variables in high-dimensional systems.

Conclusions:

• The proposed method, CIC, overcomes challenges of spurious causal interactions and latent confounders.
• The orthogonal decomposition theorem provides a robust theoretical foundation for causal inference.
• Validated effectiveness on real datasets for reconstructing biological networks and unobserved confounders.