Rapport de stage: Inférence de l'histoire démographique des chèvres par deep learning

Overview & Synopsis

As part of my internship within the VarGoats project, I first studied the structure of the project and positioned my internship subject at the core of this complex system. I analyzed the different information and communication flows to better understand the expectations and constraints. It emerged that the main challenges were financial, and the success of the project relies on effective international collaboration among the various organizations involved. Good communication is essential to facilitate knowledge sharing on this scale. My work focused on a preparatory step for future studies on goat population evolution, within the VarGoats project. I performed an imputation of sequencing data from 1,357 individuals, with the goal of conducting analyses such as the study of the demographic history of domestic goats. This process required data cleaning, followed by the use of Beagle 4.1 to compare different imputation methods. The testing protocol helped identify the best balance between imputation accuracy and computation time. It was determined that the GTGL imputation method, based on genotype likelihoods and haplotype comparisons, was the most optimal. Genotypes with a quality score below 30 were excluded to ensure better reliability. The results achieved with this method were very satisfactory, reaching a reliability level close to that of data generated by sequencing chips.