Computational and statistical analysis of copy number variants in normal and cancer genomes

AbstractAlthough the genomes from any two human individuals are more than 99.99% identical at the sequence level, some structural variation can be observed. Differences between genomes include single nucleotide polymorphism (SNP), inversion and copy number changes (gain or loss of DNA). The latter can range from submicroscopic events (CNVs, at least 1kb in size) to complete chromosomal aneuploidies. Small copy number variations have often no (lethal) consequences to the cell, but a few were associated to disease susceptibility and phenotypic variations. Larger re-arrangements (i.e. complete chromosome gain) are frequently associated with more severe consequences on health such as genomic disorders and cancer. High-throughput technologies like DNA microarrays enable the detection of CNVs in a genome-wide fashion. Since the initial catalogue of CNVs in the human genome in 2006, there has been tremendous interest in CNVs both in the context of population and medical genetics. Understanding CNV patterns within and between human populations is essential to elucidate their possible contribution to disease. But genome analysis is a challenging task; the technology evolves rapidly creating needs for novel, efficient and robust analytical tools which need to be compared with existing ones. Also, while the link between CNV and disease has been established, the relative CNV contribution is not fully understood and the predisposition to disease from CNVs of the general population has not been yet investigated.During my PhD thesis, I worked on several aspects related to CNVs. As l will report in chapter 3, ! was interested in computational methods to detect CNVs from the general population. I had access to the CoLaus dataset, a population-based study with more than 6,000 participants from the Lausanne area. All these individuals were analysed on SNP arrays and extensive clinical information were available. My work explored existing CNV detection methods and I developed a variety of metrics to compare their performance. Since these methods were not producing entirely satisfactory results, I implemented my own method which outperformed two existing methods. I also devised strategies to combine CNVs from different individuals into CNV regions.I was also interested in the clinical impact of CNVs in common disease (chapter 4). Through an international collaboration led by the Centre Hospitalier Universitaire Vaudois (CHUV) and the Imperial College London I was involved as a main data analyst in the investigation of a rare deletion at chromosome 16p11 detected in obese patients. Specifically, we compared 8,456 obese patients and 11,856 individuals from the general population and we found that the deletion was accounting for 0.7% of the morbid obesity cases and was absent in healthy non- obese controls. This highlights the importance of rare variants with strong impact and provides new insights in the design of clinical studies to identify the missing heritability in common disease.Furthermore, I was interested in the detection of somatic copy number alterations (SCNA) and their consequences in cancer (chapter 5). This project was a collaboration initiated by the Ludwig Institute for Cancer Research and involved other groups from the Swiss Institute of Bioinformatics, the CHUV and Universities of Lausanne and Geneva. The focus of my work was to identify genes with altered expression levels within somatic copy number alterations (SCNA) in seven metastatic melanoma ceil lines, using CGH and SNP arrays, RNA-seq, and karyotyping. Very few SCNA genes were shared by even two melanoma samples making it difficult to draw any conclusions at the individual gene level. To overcome this limitation, I used a network-guided analysis to determine whether any pathways, defined by amplified or deleted genes, were common among the samples. Six of the melanoma samples were potentially altered in four pathways and five samples harboured copy-number and expression changes in components of six pathways. In total, this approach identified 28 pathways. Validation with two external, large melanoma datasets confirmed all but three of the detected pathways and demonstrated the utility of network-guided approaches for both large and small datasets analysis.RésuméBien que le génome de deux individus soit similaire à plus de 99.99%, des différences de structure peuvent être observées. Ces différences incluent les polymorphismes simples de nucléotides, les inversions et les changements en nombre de copies (gain ou perte d'ADN). Ces derniers varient de petits événements dits sous-microscopiques (moins de 1kb en taille), appelés CNVs (copy number variants) jusqu'à des événements plus large pouvant affecter des chromosomes entiers. Les petites variations sont généralement sans conséquence pour la cellule, toutefois certaines ont été impliquées dans la prédisposition à certaines maladies, et à des variations phénotypiques dans la population générale. Les réarrangements plus grands (par exemple, une copie additionnelle d'un chromosome appelée communément trisomie) ont des répercutions plus grave pour la santé, comme par exemple dans certains syndromes génomiques et dans le cancer. Les technologies à haut-débit telle les puces à ADN permettent la détection de CNVs à l'échelle du génome humain. La cartographie en 2006 des CNV du génome humain, a suscité un fort intérêt en génétique des populations et en génétique médicale. La détection de différences au sein et entre plusieurs populations est un élément clef pour élucider la contribution possible des CNVs dans les maladies. Toutefois l'analyse du génome reste une tâche difficile, la technologie évolue très rapidement créant de nouveaux besoins pour le développement d'outils, l'amélioration des précédents, et la comparaison des différentes méthodes. De plus, si le lien entre CNV et maladie a été établit, leur contribution précise n'est pas encore comprise. De même que les études sur la prédisposition aux maladies par des CNVs détectés dans la population générale n'ont pas encore été réalisées.Pendant mon doctorat, je me suis concentré sur trois axes principaux ayant attrait aux CNV. Dans le chapitre 3, je détaille mes travaux sur les méthodes d'analyses des puces à ADN. J'ai eu accès aux données du projet CoLaus, une étude de la population de Lausanne. Dans cette étude, le génome de plus de 6000 individus a été analysé avec des puces SNP et de nombreuses informations cliniques ont été récoltées. Pendant mes travaux, j'ai utilisé et comparé plusieurs méthodes de détection des CNVs. Les résultats n'étant pas complètement satisfaisant, j'ai implémenté ma propre méthode qui donne de meilleures performances que deux des trois autres méthodes utilisées. Je me suis aussi intéressé aux stratégies pour combiner les CNVs de différents individus en régions.Je me suis aussi intéressé à l'impact clinique des CNVs dans le cas des maladies génétiques communes (chapitre 4). Ce projet fut possible grâce à une étroite collaboration avec le Centre Hospitalier Universitaire Vaudois (CHUV) et l'Impérial College à Londres. Dans ce projet, j'ai été l'un des analystes principaux et j'ai travaillé sur l'impact clinique d'une délétion rare du chromosome 16p11 présente chez des patients atteints d'obésité. Dans cette collaboration multidisciplinaire, nous avons comparés 8'456 patients atteint d'obésité et 11 '856 individus de la population générale. Nous avons trouvés que la délétion était impliquée dans 0.7% des cas d'obésité morbide et était absente chez les contrôles sains (non-atteint d'obésité). Notre étude illustre l'importance des CNVs rares qui peuvent avoir un impact clinique très important. De plus, ceci permet d'envisager une alternative aux études d'associations pour améliorer notre compréhension de l'étiologie des maladies génétiques communes.Egalement, j'ai travaillé sur la détection d'altérations somatiques en nombres de copies (SCNA) et de leurs conséquences pour le cancer (chapitre 5). Ce projet fut une collaboration initiée par l'Institut Ludwig de Recherche contre le Cancer et impliquant l'Institut Suisse de Bioinformatique, le CHUV et les Universités de Lausanne et Genève. Je me suis concentré sur l'identification de gènes affectés par des SCNAs et avec une sur- ou sous-expression dans des lignées cellulaires dérivées de mélanomes métastatiques. Les données utilisées ont été générées par des puces ADN (CGH et SNP) et du séquençage à haut débit du transcriptome. Mes recherches ont montrées que peu de gènes sont récurrents entre les mélanomes, ce qui rend difficile l'interprétation des résultats. Pour contourner ces limitations, j'ai utilisé une analyse de réseaux pour définir si des réseaux de signalisations enrichis en gènes amplifiés ou perdus, étaient communs aux différents échantillons. En fait, parmi les 28 réseaux détectés, quatre réseaux sont potentiellement dérégulés chez six mélanomes, et six réseaux supplémentaires sont affectés chez cinq mélanomes. La validation de ces résultats avec deux larges jeux de données publiques, a confirmée tous ces réseaux sauf trois. Ceci démontre l'utilité de cette approche pour l'analyse de petits et de larges jeux de données.Résumé grand publicL'avènement de la biologie moléculaire, en particulier ces dix dernières années, a révolutionné la recherche en génétique médicale. Grâce à la disponibilité du génome humain de référence dès 2001, de nouvelles technologies telles que les puces à ADN sont apparues et ont permis d'étudier le génome dans son ensemble avec une résolution dite sous-microscopique jusque-là impossible par les techniques traditionnelles de cytogénétique. Un des exemples les plus importants est l'étude des variations structurales du génome, en particulier l'étude du nombre de copies des gènes. Il était établi dès 1959 avec l'identification de la trisomie 21 par le professeur Jérôme Lejeune que le gain d'un chromosome supplémentaire était à l'origine de syndrome génétique avec des répercussions graves pour la santé du patient. Ces observations ont également été réalisées en oncologie sur les cellules cancéreuses qui accumulent fréquemment des aberrations en nombre de copies (telles que la perte ou le gain d'un ou plusieurs chromosomes). Dès 2004, plusieurs groupes de recherches ont répertorié des changements en nombre de copies dans des individus provenant de la population générale (c'est-à-dire sans symptômes cliniques visibles). En 2006, le Dr. Richard Redon a établi la première carte de variation en nombre de copies dans la population générale. Ces découvertes ont démontrées que les variations dans le génome était fréquentes et que la plupart d'entre elles étaient bénignes, c'est-à-dire sans conséquence clinique pour la santé de l'individu. Ceci a suscité un très grand intérêt pour comprendre les variations naturelles entre individus mais aussi pour mieux appréhender la prédisposition génétique à certaines maladies.Lors de ma thèse, j'ai développé de nouveaux outils informatiques pour l'analyse de puces à ADN dans le but de cartographier ces variations à l'échelle génomique. J'ai utilisé ces outils pour établir les variations dans la population suisse et je me suis consacré par la suite à l'étude de facteurs pouvant expliquer la prédisposition aux maladies telles que l'obésité. Cette étude en collaboration avec le Centre Hospitalier Universitaire Vaudois a permis l'identification d'une délétion sur le chromosome 16 expliquant 0.7% des cas d'obésité morbide. Cette étude a plusieurs répercussions. Tout d'abord elle permet d'effectuer le diagnostique chez les enfants à naître afin de déterminer leur prédisposition à l'obésité. Ensuite ce locus implique une vingtaine de gènes. Ceci permet de formuler de nouvelles hypothèses de travail et d'orienter la recherche afin d'améliorer notre compréhension de la maladie et l'espoir de découvrir un nouveau traitement Enfin notre étude fournit une alternative aux études d'association génétique qui n'ont eu jusqu'à présent qu'un succès mitigé.Dans la dernière partie de ma thèse, je me suis intéressé à l'analyse des aberrations en nombre de copies dans le cancer. Mon choix s'est porté sur l'étude de mélanomes, impliqués dans le cancer de la peau. Le mélanome est une tumeur très agressive, elle est responsable de 80% des décès des cancers de la peau et est souvent résistante aux traitements utilisés en oncologie (chimiothérapie, radiothérapie). Dans le cadre d'une collaboration entre l'Institut Ludwig de Recherche contre le Cancer, l'Institut Suisse de Bioinformatique, le CHUV et les universités de Lausanne et Genève, nous avons séquencés l'exome (les gènes) et le transcriptome (l'expression des gènes) de sept mélanomes métastatiques, effectués des analyses du nombre de copies par des puces à ADN et des caryotypes. Mes travaux ont permis le développement de nouvelles méthodes d'analyses adaptées au cancer, d'établir la liste des réseaux de signalisation cellulaire affectés de façon récurrente chez le mélanome et d'identifier deux cibles thérapeutiques potentielles jusqu'alors ignorées dans les cancers de la peau.

An integrative computational analysis provides evidence for FBN1-associated network deregulation in trisomy 21

Although approximately 50% of Down Syndrome (DS) patients have heart abnormalities, they exhibit an overprotection against cardiac abnormalities related with the connective tissue, for example a lower risk of coronary artery disease. A recent study reported a case of a person affected by DS who carried mutations in FBN1, the gene causative for a connective tissue disorder called Marfan Syndrome (MFS). The fact that the person did not have any cardiac alterations suggested compensation effects due to DS. This observation is supported by a previous DS meta-analysis at the molecular level where we have found an overall upregulation of FBN1 (which is usually downregulated in MFS). Additionally, that result was cross-validated with independent expression data from DS heart tissue. The aim of this work is to elucidate the role of FBN1 in DS and to establish a molecular link to MFS and MFS-related syndromes using a computational approach. To reach that, we conducted different analytical approaches over two DS studies (our previous meta-analysis and independent expression data from DS heart tissue) and revealed expression alterations in the FBN1 interaction network, in FBN1 co-expressed genes and FBN1-related pathways. After merging the significant results from different datasets with a Bayesian approach, we prioritized 85 genes that were able to distinguish control from DS cases. We further found evidence for several of these genes (47%), such as FBN1, DCN, and COL1A2, being dysregulated in MFS and MFS-related diseases. Consequently, we further encourage the scientific community to take into account FBN1 and its related network for the study of DS cardiovascular characteristics.

Statistical significance of quantitative PCR.

BACKGROUND: PCR has the potential to detect and precisely quantify specific DNA sequences, but it is not yet often used as a fully quantitative method. A number of data collection and processing strategies have been described for the implementation of quantitative PCR. However, they can be experimentally cumbersome, their relative performances have not been evaluated systematically, and they often remain poorly validated statistically and/or experimentally. In this study, we evaluated the performance of known methods, and compared them with newly developed data processing strategies in terms of resolution, precision and robustness. RESULTS: Our results indicate that simple methods that do not rely on the estimation of the efficiency of the PCR amplification may provide reproducible and sensitive data, but that they do not quantify DNA with precision. Other evaluated methods based on sigmoidal or exponential curve fitting were generally of both poor resolution and precision. A statistical analysis of the parameters that influence efficiency indicated that it depends mostly on the selected amplicon and to a lesser extent on the particular biological sample analyzed. Thus, we devised various strategies based on individual or averaged efficiency values, which were used to assess the regulated expression of several genes in response to a growth factor. CONCLUSION: Overall, qPCR data analysis methods differ significantly in their performance, and this analysis identifies methods that provide DNA quantification estimates of high precision, robustness and reliability. These methods allow reliable estimations of relative expression ratio of two-fold or higher, and our analysis provides an estimation of the number of biological samples that have to be analyzed to achieve a given precision.

Evolutionary footprint of coevolving positions in genes.

MOTIVATION: The analysis of molecular coevolution provides information on the potential functional and structural implication of positions along DNA sequences, and several methods are available to identify coevolving positions using probabilistic or combinatorial approaches. The specific nucleotide or amino acid profile associated with the coevolution process is, however, not estimated, but only known profiles, such as the Watson-Crick constraint, are usually considered a priori in current measures of coevolution. RESULTS: Here, we propose a new probabilistic model, Coev, to identify coevolving positions and their associated profile in DNA sequences while incorporating the underlying phylogenetic relationships. The process of coevolution is modeled by a 16 × 16 instantaneous rate matrix that includes rates of transition as well as a profile of coevolution. We used simulated, empirical and illustrative data to evaluate our model and to compare it with a model of 'independent' evolution using Akaike Information Criterion. We showed that the Coev model is able to discriminate between coevolving and non-coevolving positions and provides better specificity and specificity than other available approaches. We further demonstrate that the identification of the profile of coevolution can shed new light on the process of dependent substitution during lineage evolution.

Kernel-PCA data integration with enhanced interpretability

Background Nowadays, combining the different sources of information to improve the biological knowledge available is a challenge in bioinformatics. One of the most powerful methods for integrating heterogeneous data types are kernel-based methods. Kernel-based data integration approaches consist of two basic steps: firstly the right kernel is chosen for each data set; secondly the kernels from the different data sources are combined to give a complete representation of the available data for a given statistical task. Results We analyze the integration of data from several sources of information using kernel PCA, from the point of view of reducing dimensionality. Moreover, we improve the interpretability of kernel PCA by adding to the plot the representation of the input variables that belong to any dataset. In particular, for each input variable or linear combination of input variables, we can represent the direction of maximum growth locally, which allows us to identify those samples with higher/lower values of the variables analyzed. Conclusions The integration of different datasets and the simultaneous representation of samples and variables together give us a better understanding of biological knowledge.

BadiRate: Estimating Family Turnover Rates by Likelihood-Based Methods

Motivation: The comparative analysis of gene gain and loss rates is critical for understanding the role of natural selection and adaptation in shaping gene family sizes. Studying complete genome data from closely related species allows accurate estimation of gene family turnover rates. Current methods and software tools, however, are not well designed for dealing with certain kinds of functional elements, such as microRNAs or transcription factor binding sites. Results: Here, we describe BadiRate, a new software tool to estimate family turnover rates, as well as the number of elements in internal phylogenetic nodes, by likelihood-based methods and parsimony. It implements two stochastic population models, which provide the appropriate statistical framework for testing hypothesis, such as lineage-specific gene family expansions or contractions. We have assessed the accuracy of BadiRate by computer simulations, and have also illustrated its functionality by analyzing a representative empirical dataset.

Genome-wide DNA polymorphism analyses using VariScan

BACKGROUND: DNA sequence polymorphisms analysis can provide valuable information on the evolutionary forces shaping nucleotide variation, and provides an insight into the functional significance of genomic regions. The recent ongoing genome projects will radically improve our capabilities to detect specific genomic regions shaped by natural selection. Current available methods and software, however, are unsatisfactory for such genome-wide analysis. RESULTS: We have developed methods for the analysis of DNA sequence polymorphisms at the genome-wide scale. These methods, which have been tested on a coalescent-simulated and actual data files from mouse and human, have been implemented in the VariScan software package version 2.0. Additionally, we have also incorporated a graphical-user interface. The main features of this software are: i) exhaustive population-genetic analyses including those based on the coalescent theory; ii) analysis adapted to the shallow data generated by the high-throughput genome projects; iii) use of genome annotations to conduct a comprehensive analyses separately for different functional regions; iv) identification of relevant genomic regions by the sliding-window and wavelet-multiresolution approaches; v) visualization of the results integrated with current genome annotations in commonly available genome browsers. CONCLUSION: VariScan is a powerful and flexible suite of software for the analysis of DNA polymorphisms. The current version implements new algorithms, methods, and capabilities, providing an important tool for an exhaustive exploratory analysis of genome-wide DNA polymorphism data.

DnaSP v5: A software for comprehensive analysis of DNA polymorphism data

DnaSP is a software package for a comprehensive analysis of DNA polymorphism data. Version 5 implements a number of new features and analytical methods allowing extensive DNA polymorphism analyses on large datasets. Among other features, the newly implemented methods allow for: (i) analyses on multiple data files; (ii) haplotype phasing; (iii) analyses on insertion/deletion polymorphism data; (iv) visualizing sliding window results integrated with available genome annotations in the UCSC browser.

A qualitative continuous model of cellular auxin and brassinosteroid signaling and their crosstalk.

Motivation: Hormone pathway interactions are crucial in shaping plant development, such as synergism between the auxin and brassinosteroid pathways in cell elongation. Both hormone pathways have been characterized in detail, revealing several feedback loops. The complexity of this network, combined with a shortage of kinetic data, renders its quantitative analysis virtually impossible at present.Results: As a first step towards overcoming these obstacles, we analyzed the network using a Boolean logic approach to build models of auxin and brassinosteroid signaling, and their interaction. To compare these discrete dynamic models across conditions, we transformed them into qualitative continuous systems, which predict network component states more accurately and can accommodate kinetic data as they become available. To this end, we developed an extension for the SQUAD software, allowing semi-quantitative analysis of network states. Contrasting the developmental output depending on cell type-specific modulators enabled us to identify a most parsimonious model, which explains initially paradoxical mutant phenotypes and revealed a novel physiological feature.

Optimization strategies for fast detection of positive selection on phylogenetic trees.

MOTIVATION: The detection of positive selection is widely used to study gene and genome evolution, but its application remains limited by the high computational cost of existing implementations. We present a series of computational optimizations for more efficient estimation of the likelihood function on large-scale phylogenetic problems. We illustrate our approach using the branch-site model of codon evolution. RESULTS: We introduce novel optimization techniques that substantially outperform both CodeML from the PAML package and our previously optimized sequential version SlimCodeML. These techniques can also be applied to other likelihood-based phylogeny software. Our implementation scales well for large numbers of codons and/or species. It can therefore analyse substantially larger datasets than CodeML. We evaluated FastCodeML on different platforms and measured average sequential speedups of FastCodeML (single-threaded) versus CodeML of up to 5.8, average speedups of FastCodeML (multi-threaded) versus CodeML on a single node (shared memory) of up to 36.9 for 12 CPU cores, and average speedups of the distributed FastCodeML versus CodeML of up to 170.9 on eight nodes (96 CPU cores in total).Availability and implementation: ftp://ftp.vital-it.ch/tools/FastCodeML/. CONTACT: selectome@unil.ch or nicolas.salamin@unil.ch.

BadiRate: estimating family turnover rates by likelihood-based methods

Motivation: The comparative analysis of gene gain and loss rates is critical for understanding the role of natural selection and adaptation in shaping gene family sizes. Studying complete genome data from closely related species allows accurate estimation of gene family turnover rates. Current methods and software tools, however, are not well designed for dealing with certain kinds of functional elements, such as microRNAs or transcription factor binding sites. Results: Here, we describe BadiRate, a new software tool to estimate family turnover rates, as well as the number of elements in internal phylogenetic nodes, by likelihood-based methods and parsimony. It implements two stochastic population models, which provide the appropriate statistical framework for testing hypothesis, such as lineage-specific gene family expansions or contractions. We have assessed the accuracy of BadiRate by computer simulations, and have also illustrated its functionality by analyzing a representative empirical dataset.

EasyStrata: evaluation and visualization of stratified genome-wide association meta-analysis data.

The R package EasyStrata facilitates the evaluation and visualization of stratified genome-wide association meta-analyses (GWAMAs) results. It provides (i) statistical methods to test and account for between-strata difference as a means to tackle gene-strata interaction effects and (ii) extended graphical features tailored for stratified GWAMA results. The software provides further features also suitable for general GWAMAs including functions to annotate, exclude or highlight specific loci in plots or to extract independent subsets of loci from genome-wide datasets. It is freely available and includes a user-friendly scripting interface that simplifies data handling and allows for combining statistical and graphical functions in a flexible fashion. AVAILABILITY: EasyStrata is available for free (under the GNU General Public License v3) from our Web site www.genepi-regensburg.de/easystrata and from the CRAN R package repository cran.r-project.org/web/packages/EasyStrata/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Genome-wide prediction of matrix attachment regions that increase gene expression in mammalian cells.

Gene transfer in eukaryotic cells and organisms suffers from epigenetic effects that result in low or unstable transgene expression and high clonal variability. Use of epigenetic regulators such as matrix attachment regions (MARs) is a promising approach to alleviate such unwanted effects. Dissection of a known MAR allowed the identification of sequence motifs that mediate elevated transgene expression. Bioinformatics analysis implied that these motifs adopt a curved DNA structure that positions nucleosomes and binds specific transcription factors. From these observations, we computed putative MARs from the human genome. Cloning of several predicted MARs indicated that they are much more potent than the previously known element, boosting the expression of recombinant proteins from cultured cells as well as mediating high and sustained expression in mice. Thus we computationally identified potent epigenetic regulators, opening new strategies toward high and stable transgene expression for research, therapeutic production or gene-based therapies.

ENFIN--A European network for integrative systems biology.

Integration of biological data of various types and the development of adapted bioinformatics tools represent critical objectives to enable research at the systems level. The European Network of Excellence ENFIN is engaged in developing an adapted infrastructure to connect databases, and platforms to enable both the generation of new bioinformatics tools and the experimental validation of computational predictions. With the aim of bridging the gap existing between standard wet laboratories and bioinformatics, the ENFIN Network runs integrative research projects to bring the latest computational techniques to bear directly on questions dedicated to systems biology in the wet laboratory environment. The Network maintains internally close collaboration between experimental and computational research, enabling a permanent cycling of experimental validation and improvement of computational prediction methods. The computational work includes the development of a database infrastructure (EnCORE), bioinformatics analysis methods and a novel platform for protein function analysis FuncNet.

Correcting for the bias due to expression specificity improves the estimation of constrained evolution of expression between mouse and human.

MOTIVATION: Comparative analyses of gene expression data from different species have become an important component of the study of molecular evolution. Thus methods are needed to estimate evolutionary distances between expression profiles, as well as a neutral reference to estimate selective pressure. Divergence between expression profiles of homologous genes is often calculated with Pearson's or Euclidean distance. Neutral divergence is usually inferred from randomized data. Despite being widely used, neither of these two steps has been well studied. Here, we analyze these methods formally and on real data, highlight their limitations and propose improvements. RESULTS: It has been demonstrated that Pearson's distance, in contrast to Euclidean distance, leads to underestimation of the expression similarity between homologous genes with a conserved uniform pattern of expression. Here, we first extend this study to genes with conserved, but specific pattern of expression. Surprisingly, we find that both Pearson's and Euclidean distances used as a measure of expression similarity between genes depend on the expression specificity of those genes. We also show that the Euclidean distance depends strongly on data normalization. Next, we show that the randomization procedure that is widely used to estimate the rate of neutral evolution is biased when broadly expressed genes are abundant in the data. To overcome this problem, we propose a novel randomization procedure that is unbiased with respect to expression profiles present in the datasets. Applying our method to the mouse and human gene expression data suggests significant gene expression conservation between these species. CONTACT: marc.robinson-rechavi@unil.ch; sven.bergmann@unil.ch SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.