Apprentissage automatique pour le criblage de molécules sur plusieurs lignées cellulaires

par Joseph Boyd

Projet de thèse en Bio-informatique

Sous la direction de Jean-Philippe Vert et de Thomas Walter.

Thèses en préparation à Paris Sciences et Lettres , dans le cadre de Ingénierie des Systèmes, Matériaux, Mécanique, Énergétique , en partenariat avec Centre de Bio-informatique (laboratoire) et de École nationale supérieure des mines (Paris) (établissement de préparation de la thèse) depuis le 01-10-2016 .

  • Résumé

    L'objectif de cette thèse est de développer les outils algorithmiques pour permettre le criblage à haut contenu de plusieurs lignées cellulaires simultanément.

  • Titre traduit

    Machine Learning for Multi-cell-line Drug Screening

  • Résumé

    The objective of drug screening is to identify new molecules active against diseases, such as cancer. For this, a large number (typically hundreds or thousands) of experiments are performed on a cell line, which is supposed to be representative for the disease. In High Content Screening, the readout consists in images: for each individual drug, the effect is measured by fluorescence microscopy, giving thus a comprehensive description of the cellular phenotypes induced by the drug exposure. Such data sets are therefore large and complex and typically challenging to analyze. However, some diseases feature an important molecular heterogeneity, such that a single cell line can hardly be seen as a good representative for the disease. Recent publications therefore propose to systematically screen drugs on many cell lines representative for the molecular heterogeneity of the disease. These approaches are so far limited to relatively simple screens, where the drug effect is measured as an overall effect on proliferation or toxicity. The objective of this PhD project is the development of the computational tools that make HCS applicable to screens in multiple cell lines. Such a setup will allow for combining the richness and functional relevance of detailed phenotypic information derived from imaging data with the comprehensiveness regarding molecular heterogeneity brought in by the use of cell line panels.