Multi-Ancestry PRS Agent

The Multi-Ancestry PRS Agent provides AI-optimized polygenic risk score calculation designed to work across diverse ancestral populations. It addresses the critical limitation of European-biased GWAS by integrating trans-ancestry methods, improving risk prediction for underrepresented populations and enabling equitable precision medicine.

When to Use This Skill

• When calculating PRS for non-European ancestry individuals.
• For developing trans-ancestry risk prediction models.
• To reduce PRS bias across ancestral populations.
• When integrating multi-ancestry GWAS summary statistics.
• For research on PRS portability and equity.

Core Capabilities

1. Multi-Ancestry PRS: Calculate ancestry-aware polygenic scores.

1. Trans-Ancestry Optimization: Optimize weights across populations.

1. Local Ancestry Integration: Account for admixed genomes.

1. Ensemble Methods: Combine multiple PRS approaches.

1. Ancestry Calibration: Population-specific score calibration.

1. Equity Assessment: Evaluate prediction fairness across groups.

Supported Ancestries

Multi-Ancestry Methods

Workflow

1. Input: Individual genotypes, target ancestry, disease/trait.

1. Ancestry Inference: Determine genetic ancestry.

1. Method Selection: Choose optimal PRS approach.

1. Score Calculation: Compute ancestry-aware PRS.

1. Calibration: Apply population-specific calibration.

1. Risk Stratification: Categorize into risk groups.

1. Output: PRS, percentile, clinical interpretation.

Example Usage

User: "Calculate multi-ancestry coronary artery disease PRS for this admixed individual with African and European ancestry."

Agent Action:

python3 Skills/Precision_Medicine/Multi_Ancestry_PRS_Agent/calc_prs.py \
    --genotypes patient_genotypes.vcf.gz \
    --ancestry admixed_AFR_EUR \
    --local_ancestry lai_segments.bed \
    --trait coronary_artery_disease \
    --method prs_csx \
    --gwas_summary_stats eur_gwas.txt,afr_gwas.txt \
    --calibration_cohort 1kg_admixed \
    --output prs_results/

Input Requirements

Output Components

Disease-Specific Performance

AI/ML Components

PRS Optimization:

• Bayesian shrinkage (PRS-CS)
• Cross-population learning
• Neural network weight optimization

Ancestry Inference:

• Supervised classification
• Unsupervised clustering (PCA, ADMIXTURE)
• Local ancestry inference (RFMix)

Ensemble Learning:

• Stacking multiple PRS methods
• Ancestry-stratified weighting
• Uncertainty quantification

Clinical Integration

Prerequisites

• Python 3.10+
• PLINK 2.0
• PRSice-2, LDpred2, PRS-CSx
• Multi-ancestry reference panels
• GWAS summary statistics

Related Skills

• PRS_Net_Deep_Learning_Agent - Deep learning PRS
• Pharmacogenomics_Agent - Drug-gene interactions
• PopEVE_Variant_Predictor_Agent - Variant interpretation
• DiagAI_Agent - Clinical integration

Bias and Fairness

Large-Scale Initiatives

Special Considerations

1. Self-Reported Ancestry: May not match genetic ancestry
2. Admixture: Require local ancestry methods
3. Population Stratification: Careful covariate adjustment
4. Clinical Validity: Validate in target population
5. Health Equity: Consider access disparities

ESC Guidelines Integration (2025)

Limitations

Author

AI Group - Biomedical AI Platform