Reconstructing Molecular Networks by Causal Diffusion Do-Calculus Analysis with Deep Learning

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Inferring causal molecular networks is essential for understanding biological processes.
• Current methods often rely on association studies or observational causal analysis, limiting accuracy.

Purpose of the Study:

• To introduce a novel deep learning approach, Causal Diffusion Do-calculus (CDD) analysis, for inferring causal networks between molecules.
• To enhance the accuracy and generalizability of causal network inference using intervention operations within a do-calculus framework.

Main Methods:

• Developed Causal Diffusion Do-calculus (CDD) analysis, integrating intervention operations and diffusion models within a deep learning do-calculus framework.
• Applied CDD to simulated and real omics data, including UK Biobank data for causal analysis of diseases and risk factors.

Main Results:

• CDD significantly outperforms existing methods in accurately inferring gene regulatory networks.
• CDD reliably identifies disease-related genes for complex diseases, surpassing algorithms like Mendelian randomization.
• Validated CDD's effectiveness in causal analysis between diseases and potential factors across diverse populations.

Conclusions:

• CDD analysis offers a powerful and accurate method for causal network inference from observed data.
• The approach enhances the elucidation of molecular mechanisms and identification of disease-associated genes.
• CDD demonstrates superior performance and generalizability for complex biological and disease-related causal inference.