Machine learning for genomics

Le Groupe de Travail (GT) PASADENA financé par DigiCosme organise en partenariat avec les équipes BioInformatique LRI et LIX ses prochains séminaires les jeudis 14 et 28 mars 2019 au LRI - Bâtiment 650 - Salle 465 - 91400 Orsay cedex France

Programme

• Jeudi 14 mars - 14h30
  • Intervenant : Romain Menegaux (MINES ParisTech/ Institut Curie) Title : Continuous embeddings of DNA sequencing reads, and application to metagenomics
  • Résumé : We propose a new model for fast classification of DNA sequences output by next generation sequencing machines. The model, which we call fastDNA, embeds DNA sequences in a vector space by learning continuous low-dimensional representations of the k-mers it contains. We show on metagenomics benchmarks that it outperforms state-of-the-art methods in terms of accuracy and scalability.
• Jeudi 28 mars - 14h30
  • Intervenant : Blaize Hanczar (IBISC / U Evry) Title: Deep learning for phenotype prediction based on gene expression data
  • Résumé : Today, an increasing effort is put in the field of Precision Medicine to better characterize patients using high resolution technologies (also known as omics) designed to profile different facets of human biology (i.e. genomics, transcriptomics, metabolomics,…). Our contribution is about the prediction of phenotype based on gene expression data with a deep neural network. We focus on two issues: the learning with a small training set and the interpretation of the network. For the small training set problem, we propose methods based on transfer learning and semi-supervised learning. For interpretation, we backpropagate the predictions through the network in order to identify relevant genes and neurons that we associate them to biological knowledge.