Recherche

SILEX est un meta-projet collaboratif porté par l'UMR MISTEA et impliquant une équipe de développement issue d'unités partenaires avec pour thèmes le phénotypage de plantes, les bio-procédés et la viticulture. SILEX est constitué de composants logiciels mutualisés, de modèles et d'outils. SILEX aborde plusieurs aspects : mesures en-ligne, mesures hors-ligne, architecture, modèles de données et de connaissances, accès Web et traitements statistiques. Le groupe de travail associé à ce projet, se réunit toutes les semaines.

The Erdös–Rényi model of a network is simple and possesses many explicit expressions for average and asymptotic properties, but it does not fit well to real-world networks. The vertices of those networks are often structured in unknown classes (functionally related proteins or social communities) with different connectivity properties. The stochastic block structures model was proposed for this purpose in the context of social sciences, using a Bayesian approach. We consider the same model in a frequentest statistical framework. We give the degree distribution and the clustering coefficient associated with this model, a variational method to estimate its parameters and a model selection criterion to select the number of classes. This estimation procedure allows us to deal with large networks containing thousands of vertices. The method is used to uncover the modular structure of a network of enzymatic reactions.

This package contains functions for normalisation,differentially analysis of microarray data and others functions implementing recent methods developed by the Statistic and Genom Team from UMR 518 AgroParisTech/INRA Appl. Math. Comput. Sc.

Clustering the nodes of a graph allows the analysis of the topology of a network.

The stochastic block model is a clustering method based on a probabilistic model. Initially developed for binary networks it has recently been extended to valued networks possibly with covariates on the edges.

We present an implementation of a variational EM algorithm. It is written using C++, parallelized, available under a GNU General Public License (version 3), and can select the optimal number of clusters using the ICL criteria. It allows us to analyze networks with ten thousand nodes in a reasonable amount of time.

Le modèle Mixnet est un modèle de mélange pour graphes aléatoires (orientés ou non). Il vise à détecter des groupes de sommets ayant des profils de connexion similaires. L'estimation des paramètres est effectuée selon une méthode variationnelle qui permet de maximiser la vraisemblance approchée des données.

Mixnet is C++ package based on a probabilistic model for heterogeneous random graphs. This model is based on the hypothesis that real networks are made of classes of vertices which show specific connectivity patterns. Mixnet compute variationnal estimates of the parameters and a statistical criterion, ICL to select the number of classes.

This package contains functions for normalisation,differentially analysis of microarray data and others functions implementing recent methods developed by the Statistic and Genom Team from UMR 518 AgroParisTech/INRA Appl. Math. Comput. Sc.

Even if one of the major applications of two-color DNA microarray hybridizations is to detect differentially expressed genes using intensity log-ratios, single channel signals provide also useful information as absolute value measurements which allow the description of gene expression patterns. In this context, it becomes crucial to determine the set of probes that hybridize, that is for which the intensity signal is greater than a hybridization threshold to be fixed. Existing procedures are either an arbitrary thresholding or require the knowledge of a population of non-hybridized probes. In this work we present the MixThres method to determine an adaptive hybridization threshold from intensity levels of the complete set of probes hybridized on a chip. We define a hybridization threshold based on the histogram of the probe intensity values. Our procedure is divided into two steps. First the intensity distribution is estimated using mixture models. Second a hybridization threshold is defined from the components of the mixture. We validate our method on DNA tiling array and expression array data. We show that our method has a good reproducibility, its specificity is greater than 97 % and its precision of 88 %. The R package MixThres is available at http://www.agroparistech.fr/mia/outil.htm

Modèle de mélange de deux régressions linéaires permettant d'analyser simultanément plusieurs réplicats biologiques de données de ChIP-chip

2 samples comparison of Tiling Array data using mixture model approach and taking into account the dependence between probes and the structural annotation

Tiling arrays make possible a large scale exploration of the genome with high resolution. Biological questions usually addressed are either the gene expression or the detection of transcribed regions which can be investigated via transcriptomic experiments, and also the regulation of gene expression thanks to ChIP-chip experiments. In order to analyse ChIP-chip and transcriptomic data, we propose latent variable models, especially Hidden Markov Models, which are part of unsupervised classification methods. The biological features of the tiling arrays signal, such as the spatial dependence between observations along the genome and structural annotation are integrated in the model. Moreover, the models are adapted to the biological question at hand and a model is proposed for each type of experiment. We propose a mixture of regressions for the comparison of two samples, when one sample can be considered as a reference sample (ChIP-chip), and a two-dimensional Gaussian model with constraints on the variance parameter when the two samples play symmetrical roles (transcriptome). Finally, a semi-parametric modeling is considered, allowing more flexible emission distributions. With the objective of classification, we propose a false-positive control in the case of a two-cluster classification and for independent observations. Then, we focus on the classification of a set of observations forming a region of interest such as a gene. The different models are illustrated on real ChIP-chip and transcriptomic datasets coming from a NimbleGen tiling array covering the entire genome of Arabidopsis thaliana.

cghseg is an R package dedicated to the analysis of CGH profiles using segmentation models

Performs an exact Bayesian segmentation on data and returns the probabilities of breakpoints, an ICL criteria, comparison of change-point location, etc.

Performs a fast exact segmentation on data and allows for use of various cost functions.

The main goal of Multiland is to generate neutral landscapes made of several types of regions, with an exact control of the proportions occupied by each type of region. An important feature of the software is that it allows a control of the landscape fragmentation. It is intended to theoretical studies on the effect of landscape structure in applied sciences. It has been developed in the framework of the PEERLESS ANR project ``Predictive Ecological Engineering for Landscape Ecosystem Services and Sustainability".

Performs a fast exact segmentation on data and allows for use of various cost functions.

This package allows one to deal with a 2-states HMM with a fixed Gaussian distribution as emission for one state and a mixture of Gaussian for the other state. Furthermore, by computing optimal variational weights, it calculates an averaged estimator of the posterior probabilities.