Causal modeling in a multi-omic setting: insights from GAW20.

Area of Science:

• Genetics and Genomics
• Biostatistics
• Computational Biology

Background:

• Multilayered omics data availability presents opportunities and challenges for causal inference in complex diseases.
• The GAW20 Causal Modeling Working Group addressed these challenges using diverse analytical approaches.
• Focus was on identifying causal effects of genomic and epigenomic variations on lipid phenotypes.

Purpose of the Study:

• To discover novel causal relationships between genomic/epigenomic variations and lipid phenotypes.
• To validate previous findings from observational studies using causal inference methods.
• To explore bidirectional causal effects and indirect effects in multi-omics data.

Main Methods:

• Mendelian randomization (MR) with novel instrumental variable selection in methylation data.
• Structural Equation Modeling (SEM) to detect pleiotropic effects.
• Bayesian networks for causal discovery.
• A novel weighted R-squared measure for controlling outliers.

Main Results:

• MR studies identified bidirectional causal effects between CPT1A, triglycerides, RNMT, C6orf42, and high-density lipoprotein cholesterol response.
• CPT1A findings were corroborated by Bayesian network analysis.
• SEM identified pleiotropic effects, and studies estimated indirect effects of genomic variation via DNA methylation.
• A new weighted R-squared metric was proposed for robust causal inference.

Conclusions:

• GAW20 contributions demonstrate the versatility of causal inference methods in multi-omics research.
• Highlights the assumptions and strengths of different approaches (MR, SEM, Bayesian networks).
• Emphasizes the benefit of integrating multiple methods and omics layers for comprehensive disease insights.