Skip to navigation Skip to main content Skip to footer

Approved research

Deep learning prediction of complex traits

Principal Investigator: Professor Jacques Fellay
Approved Research ID: 27081
Approval date: April 1st 2017

Lay summary

The genetic architecture of complex human traits is still largely unclear, with known genetic associations only explaining a fraction of the estimated heritability. There has been no report yet describing the use of artificial intelligence for phenotypic trait prediction. We here aim at developing deep learning algorithms that can predict complex human traits from genome-wide genotyping data, starting with highly heritable but genetically poorly explained phenotypes like height and body mass index. Developing new strategies and tools that will allow the community to extract more information from large-scale genotyping data is a necessary step on the road to more personalized healthcare. We will attempt to train models that can predict complex phenotypes (e.g. height and BMI) from [1] lists of associated polymorphisms previously identified through GWAS approaches, and [2] all common human genetic variation, without a priori selection of variants. We will use different variants of Deep Neural Networks to find and quantify correlations between the SNPs and the phenotypes. Full Cohort