Skip to navigation Skip to main content Skip to footer

Approved Research

Deep Learning models for predicting phenotypic quantitative traits from high density genotyping data

Principal Investigator: Dr Eric Barrey
Approved Research ID: 96326
Approval date: March 29th 2023

Lay summary


The use of genotyping methods is becoming a routine analysis in predictive medicine (oncology), genealogy research, Gene Wide-Association Study in diseases or psychological traits,  and also in another field : animal genetic which is our research topic for production and health.  The aims of our project is develop Deep Learning (DL) or Machine Learning (ML) models  to predict  quantitative phenotypes by using the genotyping data of the individuals.

Scientific rationale

The big amount of genotyping data overreached the computation capacity of the classical methods of phenotype prediction. In order to improve the predictive models of quantitative phenotypes, we propose to apply DL/ML methods more adapted to the big data and non-linear response.


The development of these DL models needs a lot of data to be trained in a first step before being able to predict in a second step. We will implement DL/ML models  i) to generate artificial  genotypes  and phenotypes using generative adversial neural networks model (GAN) and then ii ) to combine real and artificial data to train a DL/ML models to predict quantitative phenotypes by using the genotyping data of the individuals. 


The generative model could be useful to produce artificial data genotypes-phenotypes nearly similar to real data but really anonymous that can be used for other statistical analysis, GWAS  or models.  The optimisation of these DL/ML models applied to genomics could be a determinant progress for many applications in human medicine and animal genetics.