Genomic platform specific polygenic risk scores impact breast cancer risk stratification.

BACKGROUND: At present, there is no consensus on which genotyping platform should serve as the standard for clinical polygenic risk score (PRS) implementation. Previous studies have compared the overall performance and concordance of different genotyping and sequencing technologies; however, these analyses have generally averaged the results over the whole genome. We evaluated differences in a 313-variant breast cancer PRS (PRS(313)) across genomic platforms and their impact on risk stratification. METHODS: We compare PRS(313) derived from genotyping arrays (Global Screening Array [GSA], OncoArray-500K [OncoArray], Global Diversity Array [GDA], custom Axiom_PrecipV1 array [ThermoFisher]) and low-coverage genome sequencing (lc-WGS) in 2 cell lines and 92 individuals. Probes are designed for all variants on ThermoFisher (success rate: 259/313). Sanger sequencing profiles indels. Concordance of high-risk classification (PRS(313)scoresum > 0.6) across platforms is assessed using Kappa statistics. RESULTS: In saliva samples, indel concordance with Sanger sequencing varies widely (Kappa: 0.007-1.000). PRS(313)-ThermoFisher is predictable from other platforms using linear models, despite systematic differences. Greater agreement is observed between arrays with high imputation overlap (e.g., GDA ~ GSA slope=0.986). Agreement in high-risk classification before mean correction is moderate (Fleiss's Kappa=0.552) and improves after mean correction (Kappa=0.650). Arrays with similar designs show higher agreement before mean correction (Kappa=0.745). Mean correction narrows high-risk proportions from 4-45% to 15-21%. Overall, 26 of 92 samples are classified as high risk on at least one platform, but only 7 are high risk across all. When restricting to identical variants across all platforms for PRS(313) calculation, the corresponding number of high-risk individuals are 24 and 11. CONCLUSION: Our findings demonstrate that platform-specific variability can influence PRS(313) estimates to potentially reclassify individuals around clinically relevant thresholds.