Modèles markoviens graphiques pour la fusion de données individuelles et d'interactions : application à la classification de gènes

Matthieu Vignes 1
1 MISTIS - Modelling and Inference of Complex and Structured Stochastic Systems
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, INPG - Institut National Polytechnique de Grenoble
Abstract : The research work presented in this dissertation is on keeping with the statistical integration of post-genomics data of heterogeneous kinds. Gene clustering aims at gathering genes of a living organism -modeled as a complex system- in meaningful groups according to experimental data to decipher the roles of the genes acting within biological mechanisms under study.

We based our approach on probabilistic graphical models. More specifically, we used Hidden Markov Random Fields (HMRF) that allow us to simultaneously account for gene-individual features thanks to probability distributions and network data that translate our knowledge on existing interactions between these genes through a non oriented graph.

Once the biological issues tackled are set, we describe the model we used as well as algorithmic strategies to deal with parameter estimation (namely mean field-like approximations). We then examine two specificities of the data we were faced to : the missing observation problem and the high dimensionality of this data. They lead to refinements of the model under consideration. Lastly, we present our experiments both on simulated and real Yeast data to assess the gain in using our method. In particular, our goal was to stress biologically plausible interpretations of our results.
Document type :
Theses
Mathématiques [math]. Université Joseph-Fourier - Grenoble I, 2007. Français
Liste complète des métadonnées

Cited literature [9 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00178348
Contributor : Matthieu Vignes <>
Submitted on : Thursday, November 1, 2007 - 6:48:29 PM
Last modification on : Wednesday, April 11, 2018 - 1:58:22 AM
Document(s) archivé(s) le : Tuesday, September 21, 2010 - 3:02:11 PM

Identifiers

  • HAL Id : tel-00178348, version 2

Collections

Citation

Matthieu Vignes. Modèles markoviens graphiques pour la fusion de données individuelles et d'interactions : application à la classification de gènes. Mathématiques [math]. Université Joseph-Fourier - Grenoble I, 2007. Français. 〈tel-00178348v2〉

Share

Metrics

Record views

551

Files downloads

1975