Quantifying posterior effect size distribution of susceptibility loci by common summary statistics

Area of Science:

• Statistical genetics
• Bioinformatics
• Computational biology

Background:

• Genetic association studies routinely test millions of single nucleotide polymorphisms (SNPs) for disease gene discovery.
• Recent statistical practice re-evaluations question the suitability of p-values as sole summaries of statistical evidence.
• Despite criticism, p-values contain valuable information that can be leveraged to address concerns about their limitations.

Purpose of the Study:

• To develop a new statistical method for estimating odds ratios (OR) using p-values, sample size, and standard deviation of ln(OR).
• To create a flexible approach that combines classical and Bayesian statistical principles.
• To provide direct probability statements about hypotheses for OR and resist biases from selecting top-scoring SNPs.

Main Methods:

• A novel method is presented for estimating odds ratios (OR) based on the approximate posterior distribution derived from p-values.
• The method requires only p-values, sample size, and standard deviation for ln(OR) as data summaries.
• It incorporates a flexible prior distribution for ln(OR), allowing for various shapes and representing a hybrid classical-Bayesian approach.

Main Results:

• The proposed method yields direct probability statements about hypotheses for OR, a key advantage of Bayesian approaches.
• It demonstrates resistance to biases commonly introduced by the selection of top-scoring SNPs in large-scale studies.
• The method offers greater flexibility in assumed distributions and prior forms compared to similar existing methods.

Conclusions:

• The developed method effectively utilizes p-value information for robust odds ratio estimation in genetic association studies.
• It provides a flexible and reliable tool for assessing the evidence and reliability of findings, particularly in studies involving numerous predictors.
• The approach is illustrated with interval estimates of effect size for genetic associations with lung cancer, showing broad applicability beyond OR.