Improved breast cancer risk prediction using chromosomal-scale length variation

Area of Science:

• Genomics
• Medical Genetics
• Cancer Research

Background:

• Early breast cancer detection improves survival rates.
• Current polygenic risk scores (PRS) using SNPs have limitations, especially in non-European ancestry populations.
• Chromosomal-scale length variation (CLV) offers a novel approach for genomic characterization in PRS.

Purpose of the Study:

• To develop and evaluate a breast cancer genetic risk score utilizing CLV.
• To assess the performance of this CLV-based PRS across different racial groups within the NIH All of Us dataset.
• To compare CLV-based PRS performance when trained on distinct population subgroups.

Main Methods:

• Utilized the NIH All of Us dataset, including 4,533 women with breast cancer and 44,518 controls.
• Calculated 88 CLV parameters (average log R ratios) for each individual across 22 autosomes.
• Employed machine learning algorithms to build a predictive model differentiating breast cancer cases from controls based on CLV parameters.

Main Results:

• The optimal CLV-based PRS achieved an Area Under the Curve (AUC) of 0.70 in the All of Us population.
• Individuals in the top quintile of risk scores were nine times more likely to have breast cancer compared to those in the lowest quintile.
• The CLV-based PRS demonstrated a substantial improvement over existing SNP-based PRS.

Conclusions:

• CLV-based genetic risk scoring represents a significant advancement over SNP-based polygenic risk scores for breast cancer.
• Models trained on White women generally performed better than those trained on Black women when tested on White women.
• No significant performance disparity was observed between models trained on White or Black women when tested on Black women.