Accurate Genomic Prediction of Human Height

Area of Science:

• Genetics
• Statistical Genomics
• Machine Learning

Background:

• Complex human quantitative traits are heritable but difficult to predict genetically.
• Previous methods have not fully captured the genetic basis of these traits.

Purpose of the Study:

• To construct accurate genomic predictors for complex human traits using advanced statistical methods.
• To assess the proportion of trait variance explained by common single nucleotide polymorphisms (SNPs).

Main Methods:

• Utilized high-dimensional statistics, specifically machine learning, on large-scale genomic data (UK Biobank).
• Developed predictors for height, heel bone density, and educational attainment.
• Validated predictor performance on independent datasets.

Main Results:

• Predictors explained approximately 40% (height), 20% (heel bone density), and 9% (educational attainment) of trait variance.
• Height predictions achieved a correlation of ~0.65 with actual height.
• The variance explained for height approaches the estimated common SNP heritability, suggesting capture of most SNP-based heritability.

Conclusions:

• Genomic prediction models can effectively capture a substantial portion of heritability for complex human traits.
• The developed predictors for height significantly narrow the gap between prediction R-squared and common SNP heritability.
• The identified SNPs provide insights into the genetic architecture of human height.