Development and validation of a multiancestry and multitrait polygenic risk score for lung cancer.

Zhang, Yixin; Dai, Jinglan; Gu, Pan; Zhao, Yang; Christiani, David C; Chen, Feng; Shen, Sipeng · Nat Commun · 2026

Where this comes from

Abstract

Polygenic risk scores (PRSs) quantify genetic susceptibilities, yet ancestry imbalance in genome-wide association studies (GWASs) limits the accuracy of monoracial PRSs in non-European populations. Here, we perform a multiancestry GWAS meta-analysis for lung cancer (76,953 cases and 1,886,372 controls), identifying 87 conditionally independent genome-wide significant loci, including two unreported cytobands. We use a PRS construction method, PRS-CSx, to develop a multiancestry PRS (<math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MA</mi></mrow></msub></math>) which outperforms 32 published PRSs. To enhance predictive power, we construct a multitrait PRS (<math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MT</mi></mrow></msub></math>) using CatBoost, integrating 32 cross-trait PRSs across three ancestries. Combining <math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MA</mi></mrow></msub></math> and <math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MT</mi></mrow></msub></math>, we generate <math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MAMT</mi></mrow></msub></math> and validate it in independent cohorts (OncoArray, TRICL and All of Us). <math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mrow><mi>PRS</mi></mrow><mrow><mi>MAMT</mi></mrow></msub></math> demonstrates superior discriminability in European, Asian, and African populations, improves risk stratification, and identifies approximately 10% additional lung cancer cases in the UK Biobank. Individuals with elevated PLCO<sub>m2012</sub> scores and high genetic risk exhibit a 12.64-fold higher cumulative risk than those with low scores and low genetic risk, supporting precision prevention strategies.