EndoPRS-SS: Summary-statistic based incorporation of endophenotypes to improve risk prediction

Endophenotypes are critical biological markers with genetic links to disease progression. We previously developed endoPRS, a multi-trait polygenic risk score (PRS) method that leverages endophenotype data to improve disease risk prediction. However, endoPRS requires individual-level genotype-phenotype data for model fitting, restricting its broader use. To address this, we introduce endoPRS-SS, an extension which enables PRS inference using genome-wide association study (GWAS) summary statistics and LD reference panels. Through simulation and real data analysis, we demonstrate that endoPRS-SS maintains the predictive accuracy of endoPRS, while increasing computational efficiency up to 90-fold. We also extend the endoPRS framework to support the incorporation of multiple endophenotypes simultaneously and find that multi-endophenotype PRS outperform single-endophenotype PRS. Notably, incorporating both monocyte count and executive function scores significantly improves Alzheimer's disease prediction for Europeans in the UK Biobank cohort. These extensions expand the scope of endoPRS and strengthen its utility for large-scale genetic risk assessment.