Improved Use of Small Reference Panels for Conditional and Joint Analysis with GWAS Summary Statistics

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Large-scale genomic data sharing faces practicality and confidentiality challenges, limiting public access to individual-level data.
• Genome-wide association study (GWAS) summary data are commonly shared, necessitating external reference panels for linkage disequilibrium (LD) inference.
• Existing reference panels, such as the 1000 Genomes Project European sample, have small sample sizes, leading to significant LD estimation errors and increased false positives in GWAS reanalyses.

Purpose of the Study:

• To address the inaccuracies in LD estimation stemming from small reference panels in GWAS.
• To propose an alternative covariance matrix estimator for improved association testing of single nucleotide polymorphism (SNP) groups.
• To demonstrate the performance of the proposed method compared to existing approaches using simulated and real data.

Main Methods:

• Developed an alternative estimator for the covariance matrix, drawing inspiration from multiple imputation techniques.
• Utilized simulated and real genetic datasets for evaluating the proposed method.
• Compared the performance of the new estimator against standard methods relying on small reference panels.

Main Results:

• The use of small reference panels, like the 1000 Genomes Project, leads to substantial errors in LD estimation.
• These errors significantly inflate the number of false positives in subsequent GWAS analyses.
• The proposed alternative estimator demonstrated improved performance in mitigating these issues.

Conclusions:

• Small sample sizes in reference panels pose a critical limitation for accurate GWAS reanalyses.
• The developed covariance matrix estimator offers a more reliable approach for inferring LD and conducting association tests.
• This method has the potential to reduce false positives and enhance the utility of publicly available GWAS summary data.