Disentangling molecular relationships with a causal inference test.

Area of Science:

• Genetics and Systems Biology
• Statistical Genetics
• Computational Biology

Background:

• Identifying genetic loci for complex diseases requires understanding quantitative trait loci (QTL).
• High-throughput data (genotyping, RNA expression) offer insights into disease molecular basis.
• Leveraging this data to pinpoint molecular mechanisms for QTL remains a challenge.

Purpose of the Study:

• Develop a formal statistical hypothesis test to quantify uncertainty in causal inference for molecular mediators of QTL.
• Provide a p-value to measure the confidence in a molecular species mediating the association between a locus and a quantitative trait.

Main Methods:

• Formalized causal mediation into a hypothesis test using a chain of mathematical conditions.
• Computed p-values for component conditions, including tests of linkage and conditional independence.
• Employed the Intersection-Union Test to combine statistical tests into an omnibus test.

Main Results:

• Demonstrated low Type I error in simulations with hidden variables and reactive pathways.
• Showcased comparable power to other model selection and Bayesian network reconstruction methods.
• Empirically validated favorable comparison against Bayesian network methods in yeast transcriptional regulatory network reconstruction.

Conclusions:

• Proposed a novel statistical framework formalizing causal mediation into a hypothesis test.
• Established a quantitative measure of uncertainty (p-value) for causal mediation.
• Developed a computationally accessible tool for disentangling molecular relationships and understanding QTL.