Demonstrating paths for unlocking the value of cloud genomics through cross cohort analysis

Area of Science:

• Genomics and Bioinformatics
• Population Genetics
• Data Science in Health Research

Background:

• Centralized data storage in cloud-based Trusted Research Environments (TREs) is a new paradigm for large-scale genomic projects like All of Us and UK Biobank.
• Understanding the impact of TRE attributes on cross-cohort analysis is crucial for maximizing data utility.
• Lipid measures are standard phenotypes for evaluating genomic analysis approaches due to extensive prior research.

Approach:

• Conducted a Genome-Wide Association Study (GWAS) on standard lipid measures using both meta-analysis and pooled analysis of TRE data.
• Compared the results from meta-analysis and pooled analysis against an external study to assess the correlation of identified genetic loci.
• Analyzed the unique variants identified by each approach, focusing on their prevalence across different ancestral populations.

Key Points:

• Both meta-analysis and pooled analysis showed strong correlations (R² ≈ 83-97%) with known lipid-associated loci when compared to external data.
• Meta-analysis identified 90 significant variants unique to its approach, while pooled analysis identified 64 unique significant variants.
• Approximately 20% of the unique variants in each analysis group were most prevalent in non-European, non-Asian ancestry individuals, highlighting potential ancestry-specific findings.

Conclusions:

• Technical and policy decisions in TREs can lead to divergent results in cross-cohort analyses, even when using similar data.
• Differences in analytical approaches (meta-analysis vs. pooled analysis) impact the detection of genetic variants, particularly for underrepresented ancestral groups.
• Future cross-cohort analyses must carefully consider analytical methodologies to ensure equitable discovery across diverse populations and avoid exacerbating health disparities.