GWAS genetic PCs

MODULE STATUS

Note: Module 01e is deprecated and no longer needs to be run. Its previous purpose is now covered by the PCA LD-structure QC in Module 01b. Cohorts do not need to upload or check Module 01e outputs.

Developer: Developer's team

Scripts status: Deprecated; no longer required

Prerequisite scripts: Not applicable

Data upload method: Not required

Issue report to: Haotian Tang haotian.tang@bristol.ac.uk

Module 01e Overview: Conduct GWAS of genetic PCs

GWAS commonly adjust for population stratification by using PCs as covariates. However, Grinde et al. (2024) indicates that PCs may capture not only global ancestry but also local genomic features, which has been attributed to linkage disequilibrium (LD). These PCs may not correlate with global ancestry or geography. Adjusting for such PCs in GWAS may introduce collider bias and lead to spurious associations. Excluding known high LD regions and then conducting strict LD pruning before PCA reduces local artifacts in PCs effectively, while excluding high LD regions alone is insufficient.

In Module 01b, we have generated the genetic PCs by excluding high LD regions and conducting strict LD pruning (r2 = 0.1 and window = 10000kb). Here, we respectively perform GWAS analysis for genetic PC1 to PC5, to confirm that genetic PCs are not particularly highly correlated with variants. Manhattan plots are generated with absolute beta values on the y-axis to visually inspect for strong local associations.

To run the script, use the following command:

./01e-genetic_pc_gwas.sh -c /path/to/config/file

Important

Please check Please check Manhattan plots in results/01/01e. The presence of strong local associations would suggest that genetic PCs are not appropriate for use in GWAS analyses.

Example outputs of genetic PC1 to PC5: