Mots-Clés

statistique modèle linéaire études d'association agriculture diversité génétique transcriptomique

Description

Maintaining genetic diversity is crucial for allowing agriculturally important species to adapt to changing environments driven by ongoing climate change. The domestication and selection of animal and plant species have respectively exploited only a small portion of their available genetic diversity, suggesting that a substantial part of global genetic diversity remains underexploited in breeding programs. One way to address the constraints of climate change while meeting the objectives of agroecology is to effectively identify and characterize this unexploited genetic diversity. To address this gap, the PEPR Numerical Agroecology project AgroDiv is developing cutting-edge genomics and genetics approaches to functionally characterize large-scale datasets from several agriculturally important plant and animal species.

A major goal within AgroDiv is the identification of subpopulation-specific links between gene expression and genetic variation; these subpopulations correspond to breeds (animals) or ecotypes (plants) that have often undergone strong selection for agriculturally relevant traits. In this internship, our objective is to develop a large-scale simulation study to evaluate the performance of several state-of-the-art strategies to identify subpopulation-specific associations between gene expression and genetic variants:

1. We will make use of a large-scale (n=300) set of paired whole genome sequencing (~25M genetic polymorphisms) and transcriptomic (~13k genes) data in 3 pig breeds1, generated as part of the H2020 EU GENE-SWitCH project (https://www.gene-switch.eu/). Initial work in the internship will correspond to familiarization with these data (e.g. exploratory analyses, …) as well as the use of state-of-the-art tools for the analysis of genomic and transcriptomic data (e.g., PLINK, DESeq2, RRBLUP, …).

2. We have already identified a number of candidate tools that are a promising approach to extract sub-population specific associations between gene expression and genetic variants. These include mashr2, meta-analyses as implemented in METAL3, and more straightforward strategies like linear mixed models with interaction terms or fitted independently in each sub-population. To allow for full genome-wide eQTL analyses, methods such as MatrixEqtl4 will likely be considered to allow for genome scalability.

3. To conduct a full evaluation of these different tools, a simulation strategy will be designed and will make use of the available data. It will notably include the simulation of common associations with similar or varying effect sizes across sub-populations, as well as with similar or marked differences in allelic frequency (including population-specific variants). Combined with different parameters, we aim to identify the strengths and limitations of the available modeling strategies in combination with the simulated specificity of associations.

This initial simulation study will pave the way to developing new apposite statistical approaches to efficiently identify population-specific genetic variants of interest in a large-scale omics data integration framework.