Ancestry Analysis Workflows (WDL)

This section documents the All of Us ancestry analysis workflows used to convert large callsets, identify shared high-quality sites, infer ancestry labels, and perform downstream cohort QC/relatedness analyses.

Quick Summary

• Purpose: Convert and filter large callsets, build shared high-quality site sets, infer ancestry labels, and perform post-inference sample QC.
• Primary Outputs: Per-chromosome VCFs, HQ-intersection callsets, ancestry predictions, relatedness artifacts, and interactive QC plots.

High-level Analysis Flow

Ancestry analysis is typically run in two phases:

1. Core ancestry inference preparation and prediction
2. Post-inference relatedness and outlier QC

The first phase produces ancestry labels and PCA features used by the second phase.

Phase 1: Core Ancestry Inference

Run these workflows in order:

Step	Workflow	Description	WDL
1	VDS to VCF	Converts a large VDS into per-contig full and sites-only VCF files for downstream ancestry processing.	WDL
2	Determine HQ Sites Intersection	Intersects training and input-data variants and filters both datasets to the shared high-quality site set.	WDL
3	Run Ancestry Inference	Trains and applies a PCA-based Random Forest classifier to infer ancestry labels and generate plots.	WDL

Phase 2: Post-inference Cohort QC and Relatedness

Depending on analysis goals, these workflows are used after Phase 1 outputs are available:

Step	Workflow	Description	WDL
4	Run Relatedness	Computes sample-pair relatedness and a maximal independent set of samples flagged for relatedness deconfounding.	WDL
5	Run Sample Outlier QC	Joins ancestry predictions with callset metrics and identifies ancestry-stratified sample QC outliers.	WDL
6	Run Sample Outlier QC Plotting	Joins demographics and generates interactive PC and metric visualizations for outlier QC review.	WDL

Notes

• Data dependency: run_sample_outlier_qc requires ancestry outputs from run_ancestry.
• Plotting dependency: run_sample_outlier_qc_plotting requires outputs from run_sample_outlier_qc.
• Relatedness usage: run_relatedness is commonly used to flag related samples before downstream association analyses.

Feedback

Please help us make our tools better by filing an issue in WARP; we welcome pipeline-related suggestions or questions.