Artificial Intelligence for Cross-Species Genetics in Agronomy

The genomic and functional characterisation of farm animals appears to be a lever for agroecological transition through, among other things, the identification of genotype-phenotype links. Pan-genome association studies have identified thousands of variants associated with complex agronomic traits. However, the majority of these variants have been found in non-coding genomic regions, preventing understanding of the underlying biological mechanism. Predicting molecular processes based on DNA sequence using deep learning methods is a promising approach to understanding the role of these non-coding variants. Supervised learning requires DNA sequences associated with functional data for training, which are severely limited in livestock species but available in humans. As these species are phylogenetically close, it can be hypothesised that the mechanisms regulating their genes are similar.

The project aims to :

1. evaluate different neural networks trained in humans and mice to predict functional annotations in livestock species
2. predict the impact of variants in a genomic region associated with traits of agronomic interest in order to prioritise these variants for functional validation in the laboratory.

Contact

• Noémien Maillard
• Linkedin