LiMA: Robust inference of molecular mediation from summary statistics

Area of Science:

• Genetics and bioinformatics
• Statistical genetics
• Molecular epidemiology

Background:

• Understanding molecular mechanisms of complex traits is crucial for disease intervention.
• Statistical mediation analysis, particularly Mendelian randomization, identifies causal pathways but suffers from biases due to varying sample sizes in summary statistics.
• Existing methods struggle with accurate mediation proportion estimation for numerous mediators.

Purpose of the Study:

• To introduce Likelihood-based Mediation Analysis (LiMA) for more accurate and robust molecular mediation estimation.
• To address limitations of current Mendelian randomization-based mediation methods, especially concerning measurement errors from differing sample sizes.
• To enable reliable mediation analyses across large sets of molecular mediators.

Main Methods:

• Developed Likelihood-based Mediation Analysis (LiMA) to jointly model variability in all relevant estimates.
• Employed extensive simulation studies and benchmarking to evaluate LiMA's performance against state-of-the-art methods.
• Applied LiMA to real-world data for identifying molecular mediators of cardiometabolic outcomes.

Main Results:

• LiMA demonstrated several-fold lower bias compared to existing methods.
• LiMA showed improved control of type I error rates, particularly crucial for multi-mediator analyses.
• Identified plausible metabolites (e.g., glutamate, carnitine) and proteins mediating effects of obesity risk factors on cardiometabolic outcomes.

Conclusions:

• LiMA provides a more accurate and robust framework for molecular mediation analysis.
• The method effectively accommodates variability in summary statistics of differing precision.
• LiMA has the potential to uncover novel molecular pathways underlying complex diseases, aiding targeted interventions.