Quantile Regression Models for Genetic Discovery and Phenotype Prediction

Many complex traits such as schizophrenia, Body Mass Index and various cancers have a substantial heritable component as evidenced by the large number of genetic risk variants discovered by genome-wide association studies (GWAS).

Despite these successes, a substantial gap remains between our current understanding and the ability to translate this knowledge into targeted treatment strategies.A major limitation to the conventional GWAS approaches is the reliance on oversimplified statistical models that fail to capture the complexity underlying genotype-phenotype associations.

For example, many human diseases are age-related, and many genetic associations are also age dependent.

Such underlying complexity induces heterogeneity in genetic associations and linear regression models in GWAS are not well suited to decipher such heterogeneous associations.The main goal of this project is to develop efficient statistical and scalable computational tools for GWAS and polygenic risk score (PRS) prediction based on quantile regression that will provide new insights into GWAS discoveries and PRS prediction beyond what is possible with existing methods.

Existing tools for quantile regression cannot handle the complexities of modern GWAS datasets, including ultra high-dimension and related individuals.

Therefore the project will lead to significant advances in both statistical/computational methods for GWAS, and a better understanding of genetic risk factors for complex phenotypes.