Prédiction d'expression de gènes à partir de signaux d'encombrement de la chromatine chez Fusarium graminearum
Year: 2024
Type: dissertation
Abstract: In the context of climate change, plant pathogens and pests must enhance their adaptive capacities to acclimate to recurrent stress episodes. On these timescales, modulating the epigenome appears to be an effective way for organisms to express new phenotypes that ensure their survival. Current methods for studying variations in the epigenome, including changes in chromatin accessibility and histone marks, in association with changes in gene expression, have shown limitations. Predictive performances for gene expression generally remain low. The goal of our work is to explore the complexity of epigenomic signals in relation to gene expression. This work is based on chromatin accessibility data (MAINE-seq) and expression data (RNA-seq) from the phytopathogenic fungus *Fusarium graminearum*. We propose an approach that takes into account the complexity of epigenomic signals to determine gene expression on a whole-genome scale. The study of these epigenetic variations was approached using deep learning and machine learning methods. Initial results validated the modulation of gene expression through variations in chromatin accessibility. The application of a random forest method to predict gene expression yielded interesting prediction results for a large portion of the genes. A second approach was considered, involving leveraging the common spatial supports between epigenetic and expression signals along the chromatin. This approach leads to interesting results that still need to be exploited in order to recover the expression levels commonly used in genomics.
Source: HAL
Author: Bruguet, Mathilde
Institution: Institut Agro Rennes-Angers
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W7005269964