Instrumental Variable Genetic Risk Prediction

I developed a new method to estimate the genetic correlation between key phenotypes and polygenic risk scores. This approach is both robust and privacy-protecting - it can correct for sample overlap and only requires GWAS summary-level statistics as input.

By Li Ge in Statistical Genetics Causal Inference

August 20, 2019

Rotation advisor: Qiongshi Lu, PhD.

Motivation

  • We can utilize GWAS summary statistics from major reference data sources such as UK Biobank to calculate polygenic risk scores (PRSs) for genetic risk prediction on another data set.
  • Even better, PRSs are natural instrumental variables. We can use them as proxies to investigate the genetic associations between different phenotypes.
  • However, two data set might suffer from overfitting due to sample-overlapping.
  • How can we make valid and robust inferences under sample-overlapping?

Results

  • I developed a new method to estimate the genetic correlation between key phenotypes and polygenic risk scores. This approach is both robust and privacy-protecting - it can correct for sample overlap and only requires GWAS summary-level statistics as input.

Simulation Result

Methods

See details in the slides.

Posted on:
August 20, 2019
Length:
1 minute read, 123 words
Categories:
Statistical Genetics Causal Inference
See Also: