Inflated expectations: Rare-variant association analysis using public controls.

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Publicly available sequencing datasets offer potential as controls in rare variant disease association studies.
• However, their use can increase the risk of false-positive discoveries due to unexamined factors contributing to inflated test statistics.

Purpose of the Study:

• To systematically investigate factors contributing to false-positive discoveries when using public controls in rare variant association studies.
• To quantify the inflation of statistical significance using lambda-delta-95 (λΔ95).

Main Methods:

• Leveraged public control datasets (gnomAD v2.1) and in-house sequenced datasets.
• Systematically analyzed factors including variant caller/filtering pipelines, library preparation kits, sequencers, and joint vs. separate variant-calling.
• Quantified false-positive inflation using λΔ95.

Main Results:

• Using the same variant caller and filtering pipelines for cases and controls substantially decreased test statistic inflation.
• Differences in library preparation kits and sequencers did not significantly impact the false-positive discovery rate.
• Joint versus separate variant-calling of cases and controls did not contribute to the inflation of test statistics.

Conclusions:

• Current methods inadequately adjust for high false-positive rates when using public controls.
• Risks are emphasized for rare-variant association tests using public controls when individual-level data and unified computational pipelines are inaccessible.
• Cloud-based computing offers a solution by enabling containerized pipelines to be brought to the data, minimizing these issues.