Mediation analysis of multiple mediators with incomplete omics data.

Area of Science:

• Genetics and Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Increasing use of multi-omics data (DNA, RNA, methylation, protein, metabolic profiles) to understand genotype-phenotype relationships.
• Missing omics data is a common challenge in genetic studies due to cost and sample quality issues.
• Existing mediation analysis methods often struggle with missing data and multiple mediators.

Purpose of the Study:

• To develop a robust mediation analysis framework for multi-omics data.
• To accommodate arbitrary missing data patterns across multiple omics mediators.
• To incorporate interaction effects within the mediation analysis framework.

Main Methods:

• Formulation of relationships using linear regression models for genetic variants, omics measurements, and phenotypes.
• Derivation of joint likelihood for models with two mediators, explicitly handling missing data.
• Application of computationally efficient and stable algorithms for maximum likelihood estimation.

Main Results:

• The proposed method yields unbiased and statistically efficient estimators for mediation analysis.
• Demonstrated effectiveness through comprehensive simulation studies.
• Successful application to real-world data from the Metabolic Syndrome in Men study.

Conclusions:

• The developed approach provides a powerful tool for mediation analysis with multiple omics mediators and missing data.
• The method enhances the ability to uncover complex biological pathways linking genotypes to phenotypes.
• Offers a statistically sound and computationally feasible solution for analyzing large-scale omics datasets.