Search

International audience; Methods based on univariate mixed-models are used to normalize RNA-Seq data as an initial step to detect differentially expressed (DE) genes based on the gene by experimental condition interaction term. Character state models are classically used in quantitative genetics to assess genotype by environment interactions in discrete environments. This approach, considers phenotype measurements in different environments as different traits (or character states). Thus, the interaction variance can be estimated as a function of the genetic variances and covariances of the genetic effects in the environments. In this study, we propose a multi-variate mixed model approach (i.e character state model approach) to normalize RNA-Seq data to detect DE genes as well as potential interaction variance between the genes and (i) pre and post- puberty periods and (ii) several tissues (i.e., muscle, fat, liver, uterus, ovary, pituitary gland and hypothalamus) in composite beef cattle. A total of 1,087,752 base-2 log-transformation Reads Per Kilobase of transcript per Million mapped reads (RPKM) were analyzed in the pre and post- puberty physiological states as 2 different traits in a bivariate model. The model includes the systematic effect of library (61 levels) and the random effects of gene (17,832 genes) and gene × tissue (142,656 levels). The bivariate model allowed detecting DE genes in a similar way than the univariate did (98% of DE genes in common). The interaction variance between the genes and the puberty physiological states was small (0.02) because the estimated correlation of the genes was close to unity (0.98) and the gene variances in the 2 physiological states were of similar magnitude (6.78 and 6.76 in pre and post-puberty environments, respectively). Further research is warranted to assess the optimality of a multi-variate mixed model to evaluate the interaction variance across 8 different tissues. In this second model, the log-transformed RPKM reads measured in the tissues will be considered to be different traits, while the differential expression of interest will still be between pre and post-puberty physiological states.

AU was supported by a PhD grant from EDF (Electricite de France). We are grateful to the different departments of Electricite de France for the financial support of the present study: EDF -Recherche et Developpement, Clamart especially Dr Mathieu Le Brun and Laurence Tissote EDF- Unite of Production Mediterranee especially Dr Julie Mosseri and EDF Centre dIngenierie Hydraulique Technolac Chambery especially Dr Agnes Barillier and Frederic Jacob. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.; International audience; Analyses of high-throughput transcriptome sequences of non-model organisms are based on two main approaches: de novo assembly and genome-guided assembly using mapping to assign reads prior to assembly. Given the limits of mapping reads to a reference when it is highly divergent, as is frequently the case for non-model species, we evaluate whether using blastn would outperform mapping methods for read assignment in such situations (> 15% divergence). We demonstrate its high performance by using simulated reads of lengths corresponding to those generated by the most common sequencing platforms, and over a realistic range of genetic divergence (0% to 30% divergence). Here we focus on gene identification and not on resolving the whole set of transcripts (i.e. the complete transcriptome). For simulated datasets, the transcriptome-guided assembly based on blastn recovers 94.8% of genes irrespective of read length at 0% divergence; however, assignment rate of reads is negatively correlated with both increasing divergence level and reducing read lengths. Nevertheless, we still observe 92.6% of recovered genes at 30% divergence irrespective of read length. This analysis also produces a categorization of genes relative to their assignment, and suggests guidelines for data processing prior to analyses of comparative transcriptomics and gene expression to minimize potential inferential bias associated with incorrect transcript assignment. We also compare the performances of de novo assembly alone vs in combination with a transcriptome-guided assembly based on blastn both via simulation and empirically, using data from a cyprinid fish species and from an oak species. For any simulated scenario, the transcriptome-guided assembly using blastn outperforms the de novo approach alone, including when the divergence level is beyond the reach of traditional mapping methods. Combining de novo assembly and a related reference transcriptome for read assignment also addresses the bias/error in contigs caused by the dependence on a related reference alone. Empirical data corroborate these findings when assembling transcriptomes from the two non-model organisms: Parachondrostoma toxostoma (fish) and Quercus pubescens (plant). For the fish species, out of the 31,944 genes known from D. rerio, the guided and de novo assemblies recover respectively 20,605 and 20,032 genes but the performance of the guided assembly approach is much higher for both the contiguity and completeness metrics. For the oak, out of the 29,971 genes known from Vitis vinifera, the transcriptome-guided and de novo assemblies display similar performance, but the new guided approach detects 16,326 genes where the de novo assembly only detects 9,385 genes.

As gene expression measurement technology is shifting from microarrays to sequencing, the statistical tools available for their analysis must be adapted since RNA-seq data are measured as counts. Recently, it has been proposed to tackle the count nature of these data by modeling log-count reads per million as continuous variables, using nonparametric regression to account for their inherent heteroscedasticity. Adopting such a framework, we propose tcgsaseq, a principled, model-free and efficient top-down method for detecting longitudinal changes in RNA-seq gene sets. Considering gene sets defined a priori, tcgsaseq identifies those whose expression vary over time, based on an original variance component score test accounting for both covariates and heteroscedasticity without assuming any specific parametric distribution for the transformed counts. We demonstrate that despite the presence of a nonparametric component, our test statistic has a simple form and limiting distribution, and both may be computed quickly. A permutation version of the test is additionally proposed for very small sample sizes. Applied to both simulated data and two real datasets, the proposed method is shown to exhibit very good statistical properties, with an increase in stability and power when compared to state of the art methods ROAST, edgeR and DESeq2, which can fail to control the type I error under certain realistic settings. We have made the method available for the community in the R package tcgsaseq., 23 pages, 6 figures, typo corrections & acceptance acknowledgement

International audience; The development of High Throughput Sequencing (HTS) for RNA profiling (RNA-seq) has shed light on the diversity of transcriptomes. While RNA-seq is becoming a de facto standard for monitoring the population of expressed transcripts in a given condition at a specific time, processing the huge amount of data it generates requires dedicated bioinformatics programs. Here, we describe a standard bioinformatics protocol using state-of-the-art tools, the STAR mapper to align reads onto a reference genome, Cufflinks to reconstruct the transcriptome, and RSEM to quantify expression levels of genes and transcripts. We present the workflow using human transcriptome sequencing data from two biological replicates of the K562 cell line produced as part of the ENCODE3 project.

International audience; Next-generation sequencing technologies now constitute a method of choice to measure gene expression. Data to analyze are read counts, commonly modeled using negative binomial distributions. A relevant issue associated with this probabilistic framework is the reliable estimation of the overdispersion parameter, reinforced by the limited number of replicates generally observable for each gene. Many strategies have been proposed to estimate this parameter, but when differential analysis is the purpose, they often result in procedures based on plug-in estimates, and we show here that this discrepancy between the estimation framework and the testing framework can lead to uncontrolled type-I errors. Instead, we propose a mixture model that allows each gene to share information with other genes that exhibit similar variability. Three consistent statistical tests are developed for differential expression analysis. We show through a wide simulation study that the proposed method improves the sensitivity of detecting differentially expressed genes with respect to the common procedures, since it reaches the nominal value for the type-I error, while keeping elevate discriminative power between differentially and not differentially expressed genes. The method is finally illustrated on prostate cancer RNA-Seq data.

International audience; In the past 5 years, RNA-Seq has become a powerful tool in transcriptome analysis even though computational methods dedicated to the analysis of high-throughput sequencing data are yet to be standardized. It is, however, now commonly accepted that the choice of a normalization procedure is an important step in such a process, for example in differential gene expression analysis. The present article highlights the similarities between three normalization methods: TMM from edgeR R package, RLE from DESeq2 R package, and MRN. Both TMM and DESeq2 are widely used for differential gene expression analysis. This paper introduces properties that show when these three methods will give exactly the same results. These properties are proven mathematically and illustrated by performing in silico calculations on a given RNA-Seq data set.

International audience; Les données d'expression issues du séquençage haut-débit (RNA-seq) sont des données de comptage très hétérogènes. Il est naturel de les représenter par des modèles basés sur des lois discrètes comme la loi de Poisson ou la loi binomiale négative. Mais des transformations simples des données peuvent permettre de se ramener à des modèles plus répandus fondés sur des lois gaussiennes. Nous montrons comment comparer objectivement les vraisemblances de ces modèles travaillant sur des données différentes. Nous nous focalisons pour mener ces comparaisons sur des problèmes de classification où les mélanges de Poisson et gaussiens peuvent etre mis en compétition.; High-throughput transcriptome sequencing data (RNA-seq) are made up of highly heterogeneous counts. Although they are often modeled with discrete distributions, including the Poisson and negative binomial distributions, Gaussian models on transformed data could alternatively be considered. We show how the likelihood of these different models can be objectively compared. We focus attention on the problem of clustering gene profiles, where Poisson mixtures on count data are compared with Gaussian mixtures on transformed data.

This thesis gathers methodologicals contributions to the statistical analysis of next-generation high-throughput transcriptome sequencing data (RNA-seq). RNA-seq data are discrete and the number of samples sequenced is usually small due to the cost of the technology. These two points are the main statistical challenges for modelling RNA-seq data.The first part of the thesis is dedicated to the co-expression analysis of RNA-seq data using model-based clustering. A natural model for discrete RNA-seq data is a Poisson mixture model. However, a Gaussian mixture model in conjunction with a simple transformation applied to the data is a reasonable alternative. We propose to compare the two alternatives using a data-driven criterion to select the model that best fits each dataset. In addition, we present a model selection criterion to take into account external gene annotations. This model selection criterion is not specific to RNA-seq data. It is useful in any co-expression analysis using model-based clustering designed to enrich functional annotation databases.The second part of the thesis is dedicated to network inference using graphical models. The aim of network inference is to detect relationships among genes based on their expression. We propose a network inference model based on a Poisson distribution taking into account the discrete nature and high inter sample variability of RNA-seq data. However, network inference methods require a large number of samples. For Gaussian graphical models, we propose a non-asymptotic approach to detect relevant subsets of genes based on a block-diagonale decomposition of the covariance matrix. This method is not specific to RNA-seq data and reduces the dimension of any network inference problem based on the Gaussian graphical model., Cette thèse regroupe des contributions méthodologiques à l'analyse statistique des données issues des technologies de séquençage du transcriptome (RNA-seq). Les difficultés de modélisation des données de comptage RNA-seq sont liées à leur caractère discret et au faible nombre d'échantillons disponibles, limité par le coût financier du séquençage. Une première partie de travaux de cette thèse porte sur la classification à l'aide de modèle de mélange. L'objectif de la classification est la détection de modules de gènes co-exprimés. Un choix naturel de modélisation des données RNA-seq est un modèle de mélange de lois de Poisson. Mais des transformations simples des données permettent de se ramener à un modèle de mélange de lois gaussiennes. Nous proposons de comparer, pour chaque jeu de données RNA-seq, les différentes modélisations à l'aide d'un critère objectif permettant de sélectionner la modélisation la plus adaptée aux données. Par ailleurs, nous présentons un critère de sélection de modèle prenant en compte des informations biologiques externes sur les gènes. Ce critère facilite l'obtention de classes biologiquement interprétables. Il n'est pas spécifique aux données RNA-seq. Il est utile à toute analyse de co-expression à l'aide de modèles de mélange visant à enrichir les bases de données d'annotations fonctionnelles des gènes. Une seconde partie de travaux de cette thèse porte sur l'inférence de réseau à l'aide d'un modèle graphique. L'objectif de l'inférence de réseau est la détection des relations de dépendance entre les niveaux d'expression des gènes. Nous proposons un modèle d'inférence de réseau basé sur des lois de Poisson, prenant en compte le caractère discret et la grande variabilité inter-échantillons des données RNA-seq. Cependant, les méthodes d'inférence de réseau nécessitent un nombre d'échantillons élevé.Dans le cadre du modèle graphique gaussien, modèle concurrent au précédent, nous présentons une approche non-asymptotique pour sélectionner des sous-ensembles de gènes pertinents, en décomposant la matrice variance en blocs diagonaux. Cette méthode n'est pas spécifique aux données RNA-seq et permet de réduire la dimension de tout problème d'inférence de réseau basé sur le modèle graphique gaussien.

Background: Change point problems arise in many genomic analyses such as the detection of copy number variations or the detection of transcribed regions. The expanding Next Generation Sequencing technologies now allow to locate change points at the nucleotide resolution. Results: Because of its complexity which is almost linear in the sequence length when the maximal number of segments is constant, and as its performance had been acknowledged for microarrays, we propose to use the Pruned Dynamic Programming algorithm for Seq-experiment outputs. This requires the adaptation of the algorithm to the negative binomial distribution with which we model the data. We show that if the dispersion in the signal is known, the PDP algorithm can be used, and we provide an estimator for this dispersion. We describe a compression framework which reduces the time complexity without modifying the accuracy of the segmentation. We propose to estimate the number of segments via a penalized likelihood criterion. We illustrate the performance of the proposed methodology on RNA-Seq data. Conclusions: We illustrate the results of our approach on a real dataset and show its good performance. Our algorithm is available as an R package on the CRAN repository.

International audience; In recent years, RNA-Seq technologies became a powerful tool for transcriptome studies. However, computational methods dedicated to the analysis of high-throughput sequencing data are yet to be standardized. In particular, it is known that the choice of a normalization procedure leads to a great variability in results of differential gene expression analysis. The present study compares the most widespread normalization procedures and proposes a novel one aiming at removing an inherent bias of studied transcriptomes related to their relative size. Comparisons of the normalization procedures are performed on real and simulated data sets. Real RNA-Seq data sets analyses, performed with all the different normalization methods, show that only 50% of significantly differentially expressed genes are common. This result highlights the influence of the normalization step on the differential expression analysis. Real and simulated data sets analyses give similar results showing 3 different groups of procedures having the same behavior. The group including the novel method named "Median Ratio Normalization" (MR N) gives the lower number of false discoveries. Within this group the MR N method is less sensitive to the modification of parameters related to the relative size of transcriptomes such as the number of down- and upregulated genes and the gene expression levels. The newly proposed MR N method efficiently deals with intrinsic bias resulting from relative size of studied transcriptomes. Validation with real and simulated data sets confirmed that MR N is more consistent and robust than existing methods.