Leveraging a surrogate outcome to improve inference on a partially missing target outcome

Area of Science:

• Genetics
• Statistical Genomics
• Bioinformatics

Background:

• Sample size disparities exist across tissues in the Genotype-Tissue Expression (GTEx) project, limiting power for tissue-specific expression quantitative trait loci (eQTL) detection in undersampled tissues like the substantia nigra (SSN).
• Inaccessible tissues often have fewer samples compared to accessible ones, posing a significant challenge for comprehensive genetic analyses.

Purpose of the Study:

• To introduce Surrogate Phenotype Regression Analysis (Spray), a novel statistical method designed to enhance eQTL detection in tissues with limited sample sizes.
• To leverage information from correlated surrogate outcomes (e.g., blood expression) to improve inference on partially missing target outcomes (e.g., SSN expression).

Main Methods:

• Spray employs a bivariate regression framework to jointly model target and surrogate outcomes, treating unobserved values as missing data.
• An expectation conditional maximization algorithm is implemented for estimation in the presence of bilateral outcome missingness.
• The method estimates the same association parameter as standard eQTL mapping and controls Type I error rates.

Main Results:

• Spray increases estimation precision and improves statistical power for eQTL detection compared to marginal modeling of the target outcome.
• The method demonstrates effectiveness both analytically and empirically through simulations and analysis of GTEx data.
• Spray maintains valid statistical inference even when target and surrogate outcomes are uncorrelated.

Conclusions:

• Surrogate Phenotype Regression Analysis (Spray) offers a powerful approach to overcome sample size limitations in eQTL studies.
• This method enhances the ability to identify tissue-specific genetic associations in undersampled human tissues.
• Spray represents a significant advancement in statistical genetics for leveraging available data more effectively.