Quantifying interventional causality by knockoff operation.

Area of Science:

• Computational Biology
• Systems Biology
• Network Science

Background:

• Causal inference is vital for understanding complex biological mechanisms.
• Inferring causality in biological networks computationally remains a significant challenge.

Purpose of the Study:

• To introduce knockoff conditional mutual information (KOCMI), a novel criterion for inferring interventional direct causality.
• To enable causal inference without prior knowledge of network structure, using various data types.

Main Methods:

• KOCMI utilizes a knockoff operation as a virtual intervention on variables.
• It estimates distributional invariance before and after the virtual intervention to identify causality.
• The method is applicable to both time-independent and time-series data, and networks with loops.

Main Results:

• KOCMI accurately quantifies causal relationships, even in complex networks with feedback loops.
• The method demonstrates theoretical consistency with do-calculus but without its structural prerequisites.
• KOCMI exhibits superior performance compared to existing methods on benchmark and real-world datasets.

Conclusions:

• KOCMI is a powerful, theoretically sound, and experimentally validated tool for interventional causality inference.
• It offers a robust approach to uncovering causal mechanisms in biological systems.
• The method advances computational biology by addressing key challenges in network causality.