Improving DirectLiNGAM for high-dimensional microbiome data: roots screening and eBIC based model selection.

Area of Science:

• Systems Biology
• Microbiome Research
• Causal Inference

Background:

• Gut microbiome research faces challenges in identifying causal relationships from high-dimensional, sample-limited observational data.
• Complex interactions within the gut microbiome significantly influence host health and disease.
• Existing causal discovery methods often struggle with the unique characteristics of microbiome data.

Purpose of the Study:

• To enhance the DirectLiNGAM algorithm for improved causal discovery in gut microbiome research.
• To introduce methodological improvements for practical application of causal inference in complex biological systems.
• To extend the applicability of LiNGAM-based approaches to microbiome data analysis.

Main Methods:

• Developed two complementary extensions to the DirectLiNGAM algorithm.
• Implemented prior knowledge extraction through roots screening.
• Integrated the extended Bayesian Information Criteria (BIC) for model selection.
• Validated the methodology using synthetic datasets and a real biological dataset.

Main Results:

• The proposed extensions successfully enhance DirectLiNGAM's applicability to microbiome data.
• Numerical experiments demonstrated the robustness and effectiveness of the improved methodology.
• The approach facilitates more accurate causal discovery in complex, high-dimensional biological systems.
• Successful application on a real biological dataset confirmed practical utility.

Conclusions:

• The enhanced DirectLiNGAM methodology provides a powerful tool for causal discovery in gut microbiome research.
• These improvements address key limitations of applying causal inference to complex biological data.
• The study supports the broader adoption of LiNGAM-based methods in systems biology and related fields.