Skip to navigation Skip to main content Skip to footer

Approved research

Influence of genetic architecture on polygenic prediction

Principal Investigator: Dr Wen Huang
Approved Research ID: 56579
Approval date: April 30th 2020

Lay summary

Prediction of phenotypes from genotypes is an important goal of genetics research. Even with the success in the last decade in advanced statistical and computational methodologies, our ability to make accurate predictions remain limited. The UKBB dataset provides a venue where both genotype and phenotype data are available for a large number of individuals so methods can be developed and reliably tested. To do so, we plan to work in three related areas: First, genes and sequence variants, the working units in genetics, do not work in isolation and must cooperate with each other to influence phenotypes. However, most current methods deal with genes as if they work independently from each other. The first area of our research will be to characterize, for the different phenotypes available in the UKBB, to what extend do genes and DNA sequence differences interact with each other. We will further determine how the different magnitudes of interactions may affect how well we are able to make predictions of phenotypes from genotypes alone, especially when such tasks are performed between divergent populations. Second, for a statistical model to perform well in prediction, it must take into account how the models and parameters differ between different contexts, including genetic and environmental contexts. However, current methods do not do this very well. We will develop methods to explicitly take into account the differences in genetic effects between different contexts and test their performance in genetic prediction. Third, present methodologies specify statistical models in a data driven manner and do not explicitly take into account prior knowledge on the biology of the phenotypes. There is now increasing knowledge on the biology of certain phenotypes, including how genes are regulated and/or interact with each other to generate certain biological responses. We will develop methods to specify statistical models that are biologically sensible and use them to make predictions. In summary, the proposed project will efficiently make use of the UKBB data and enhance our ability to predict phenotypes from genotypes, including many disease risks and health related traits, benefiting public health.