Approved Research

Capturing genotype-to-phenotype relationships for complex phenotypes using neural network based statistical models that leverage phenotype-phenotype relationships.

Lay summary

Genetic data is more available than ever before. Consistently, we now see many studies relating genetic factors to human traits. However, these studies frequently only predict poorly. The methodology often used to relate genes and traits does not allow for the possibility of complex interactions between genes, other genes, and environments to impact traits. Here we aim to test our new neural-network-based framework for creating such genotype-to-trait maps. Our approach allows for interactions between both genes and environments and captures correlations among phenotypes. If our method produces more accurate predictions of traits than conventional analysis, we will determine the fraction of that prediction increase that is attributable to gene-gene, potentially gene-environment, and dominance effects. An understanding of the prevalence and importance of these phenomena in determining human traits will improve our understanding of the basic drivers of human trait variation.