Stephen Hsu

The paper develops population-specific polygenic risk scores for Han Chinese using genomic and electronic health record data, showing strong predictive performance for multiple diseases.

The paper introduces blockLASSO, a method for training polygenic scores that is computationally efficient and effective across multiple phenotypes and ancestry groups.

The paper evaluates sparse machine learning algorithms for polygenic prediction, showing LASSO performs well and provides projections for new biobanks.

The paper constructs a polygenic health index using disease risk scores and validates it in the UK Biobank, finding higher scores correlate with reduced disease risk and longer life expectancy.

The paper trains machine learning models to predict blood/urine biomarkers from SNPs and disease risk from biomarkers using UK Biobank data, achieving high accuracy for several diseases.

The paper validates polygenic risk scores by comparing disease risk and trait prediction accuracy between genetic siblings and non-siblings.

The paper analyzes genetic variants used in disease risk predictors and finds that most SNPs are outside coding regions, suggesting exome data alone cannot capture heritability for complex traits.

This study constructs polygenic risk scores for multiple complex diseases using UK Biobank data and LASSO regression, achieving AUCs between 0.58 and 0.71, with improvements when incorporating age, sex, and larger case populations.

The paper constructs genomic predictors for human height, heel bone density, and educational attainment using machine learning, capturing significant variance and resolving part of the missing heritability problem.