2023Activity reportProject-TeamPLEIADE

3.1 Research: A Geometric View of Diversity

Data analysis algorithms and tools must be revisited and scaled up. We mobilize both distributed algorithms and new algorithms, like random projection or column selection methods, to build point clouds in Euclidean spaces from massive data sets, and thus overcome the cubic complexity of computation of eigenvectors and eigenvalues of very large dense matrices. We also link distance geometry 51 with convex optimization procedures through matrix completion 34, 37.

Intercalibration: There is a considerable difference between supervised and unsupervised clustering: in supervised clustering, the result for an item $i$ is independent from the result for an item $j\ne i$, whereas in unsupervised clustering, the result for an item $i$ (e.g. the cluster it belongs to, and its composition) depends on nearby items $j\ne i$. Which means that the result may change if some items are added to or subtracted from the sample. This raises the more global problem of how to merge two studies to yield a more comprehensive view of biodiversity?

See 44 for some of our recent work linking the distance geometry problem, nonlinear mapping, and weighted least-squares scaling.

3.2 Research: Community-scale Metabolic and Omics-based Modeling

Metabolism can be abstracted into sets of metabolic reactions, associated to the genome through gene-protein relationships, and connecting substrates to products thereby forming metabolic networks. Genome-scale metabolic networks (GSMNs) contain all the reactions predicted to occur in a organism according to its genomic contents. Combined with additional knowledge on the system, possibly other –omics data and mathematical models, GSMNs are used to predict the behaviour of an organism or a community of organisms.

A widely-used mathematical formalism for modeling GSMNs is constraint-based modeling, among which flux balance analysis (FBA) is the main representative 53. Such methods permit a quantitative prediction of activity fluxes in metabolic networks while optimising an objective function and assuming steady-state of the system.

The emergent metabolism of microbial communities can also be qualitatively modeled using a boolean approximation of metabolic dynamics8. In this approach the behavior of the system is described by logical rules that activate a given reaction as soon as its substrates become available; numerical parameters such as stochiometry or enzyme kinetics are ignored in favor of graph topology and paths. The advantage is that such qualitative models, unlike quantitative methods such as FBA, do not require the assumption that the system is at steady-state and can model systems where cells are constantly growing or constantly reproducing.

Network expansion, introduced in 43 as a recursive traversal of the structure of a metabolic graph, lends itself to concise definition using answer set programming (ASP) 49 and thus to efficient implementation using SAT solvers 48. In practice, using ASP for metabolic modeling makes it possible to define both the activation of metabolic reactions in different conditions, and the constraints and optimizations needed to find solutions in a combinatorically large state space.

We focus in particular on the key question of determining minimal communities, subsets of the organisms present in an environment that are sufficient to reproduce a chosen behavior 45. The methodological goal here is to identify key species in a community through use of ASP to rapidly explore the search space and thus, through heuristic resolution of combinatorial problems, provide the guarantees an exhaustive search with a greatly reduced computational cost 4.

Functional and taxonomic diversities, beyond intrinsic specificities encoded in the genetic material, are also strongly shaped by their environment. Spatial nutritional niches, microbial interactions and abiotic constraints lead to complex spatial structures in the microbial community that impact its overall dynamics. PDE-based models of the microbiota in its environment allow including in the model these multiple mechanisms in order to decipher their influence on the community faith.

The main methodological developments in this area are related to mathematical modeling (in particular the correct level of simplification in the multi-physics description of the microbial environment), model simplification (asymptotic approximation), inference from multi-omics data (including dimension reduction, statistical learning) and numerical developments (in particular fast approximation of metabolic models with machine learning methods). Strong interactions with community-scale metabolic models are sought, specially for multi-omics inference and knowledge-based machine learning constraints.

The goal is to achieve accurate models of microbial communities that could be used as digital twins of controlled experiments in microbial ecology. Culturomic facilities allow for the acquisition of multi-omics time-series data in controlled conditions, which can be used to build and fit population dynamics models, and can be used in turn to explore numerically biological assumptions and to help in experimental planning and data analysis.

3.3 Research: Bioinformatics, Genomes, and Knowledge Management

The heterogenous data generated in computational molecular biology and ecology are distinguished not only by their volume, but by the richness of the many levels of interpretation that biologists create. The same nucleic acid sequence can be seen as a molecule with a structure, a sequence of base pairs, a collection of genes, an allele, or a molecular fingerprint. To extract the maximum benefit from this treasure trove we must organize the knowledge in ways that facilitate extraction, analysis, and inference. Our focus has been on the efficient representation of relations between biological objects and operations on those representations, in particular heuristic analyses and logical inference.

Pleiade developes applications in comparative genomics of related organisms, using novel mathematical tools for representing compactly, at different scales of difference, comparisons between related genomes. New methods based on distance geometry will refine these comparisons. Compact representations can be stored, exchanged, and combined. They form the basis of novel simultaneous genome annotation methods, that can be linked directly to abductive inference methods for building functional models of the organisms and their communities.

Since a goal of Pleiade is to integrate diversity throughout the analysis process, it is necessary to incorporate diversity as a form of knowledge that can be stored in a knowledge base. Diversity can be represented using various compact representations, such as trees and quotient graphs storing nested sets of relations. Extracting structured representations and logical relations from integrated knowledge bases requires domain-specific query methods that can express forms of diversity.

4.1 Molecular based systematics and taxonomy

Defining and recognizing the myriads of species occuring in the biosphere has been the focus of phenomenal energy over the past centuries and remains a major goal of Natural History. It is an iconic paradigm in pattern recognition (clustering has coevolved with numerical taxonomy many decades ago). Developments in evolution and molecular biology, as well as in data analysis, have over the past decades enabled a profound revolution, where species can be delimited and recognized by data analysis of sequences. We aim at proposing new tools, in the framework of E-science, which make possible ($i$) better exploration of the diversity in a given clade, and ($ii$) assignment of a place in these patterns for new, unknown organisms, using information provided by sets of sequences. This will require investment in data analysis, machine learning, and pattern recognition to deal with the volumes of data and their complexity.

One example of this project is about the diversity of trees in Amazonian forest, in collaboration with botanists in French Guiana. Protists (unicellular Eukaryots) are by far more diverse than plants, and far less known. Molecular exploration of Eukaryotes diversity is nowadays a standard in biodiversity studies. Data are available, through metagenomics, as an avalanche and make molecular diversity enter the domain of Big Data. Hence, an effort will be invested, in collaboration with other INRIA teams (GenScale, HiePACS) for porting to HPC algorithms of pattern recognition and machine learning, or distance geometry, for these tools to be available as well in metagenomics. This will be developed first on diatoms (unicellular algae) in collaboration with INRAE team at Thonon and University of Uppsala, on pathogens of tomato and grapewine, within an existing network, and on bacterial communities, in collaboration with University of Pau. For the latter, the studies will extend to correlations between molecular diversity and sets of traits and functions in the ecosystem.

4.2 Genome and transcriptome annotation

Sequencing genomes and transcriptomes provides a picture of how a biological system can function, or does function under a given physiological condition. Simultaneous sequencing of a group of related organisms is now a routine procedure in biological laboratories for studying a behavior of interest, and provides a marvelous opportunity for building a comprehensive knowledge base of the relations between genomes1, 12. Key elements in mining these relations are: classifying the genes in related organisms and the reactions in their metabolic networks, recognizing the patterns that describe shared features, and highlighting specific differences.

Pleiade develops applications in comparative genomics of related organisms, using new mathematical tools for representing compactly, at different scales of difference, comparisons between related genomes. New methods based on computational geometry refine these comparisons. Compact representations can be stored, exchanged, and combined. They will form the basis of new simultaneous genome annotation methods, linked directly to abductive inference methods for building functional models of the organisms and their communities.

Our ambition in biotechnology is to permit the design of synthetic or genetically selected organisms at an abstract level, and guide the modification or assembly of a new genome. Our effort is focused on two main applications: genetic engineering and synthetic biology of oil-producing organisms (biofuels in CAER, palm oils), and improving and selecting starter microorganisms used in winemaking (collaboration with the ISVV and the BioLaffort company).

4.3 Community ecology

Community assembly models how species can assemble or diassemble to build stable or metastable communities. It has grown out of inventories of countable organisms. Using metagenomics one can produce molecular based inventories at rates never reached before. Most communities can be understood as pathways of carbon exchange, mostly in the form of sugar, between species. Even a plant cannot exist without carbon exchange with its rhizosphere. Two main routes for carbon exchange have been recognized: predation and parasitism. In predation, interactions–even if sometimes dramatic–may be loose and infrequent, whereas parasitism requires what Claude Combes has called intimate and sustainable interactions 38. About one decade ago, some works 56 have proposed a comprehensive framework to link the studies of biodiversity with community assembly. This is still incipient research, connecting community ecology and biogeography.

We aim at developping graph-based models of co-occurence between species from NGS inventories in metagenomics, i.e. recognition of patterns in community assembly, and as a further layer to study links, if any, between diversity at different scales and community assemblies, starting from current, but oversimplified theories, where species assemble from a regional pool either randomly, as in neutral models, or by environmental filtering, as in niche modeling. We propose to study community assembly as a multiscale process between nested pools, both in tree communities in Amazonia, and diatom communities in freshwaters. This will be a step towards community genomics, which adds an ecological flavour to metagenomics.

Next-generation sequencing technologies are now an essential tool in population and community genomics, either for making evolutionary inferences or for developing SNPs for population genotyping analyses. Two problems are highlighted in the literature related to the use of those technologies for population genomics: variable sequence coverage and higher sequencing error in comparison to the Sanger sequencing technology. Methods are developed to develop unbiased estimates of key parameters, especially integrating sequencing errors 52. An additional problem can be created when sequences are mapped on a reference sequence, either the sequenced species or an heterologous one, since paralogous genes are then considered to be the same physical position, creating a false signal of diversity 47. Several approaches were proposed to correct for paralogy, either by working directly on the sequences issued from mapped reads 47 or by filtering detected SNPs. Finally, an increasingly popular method (RADseq) is used to develop SNP markers, but it was shown that using RADseq data to estimate diversity directly biases estimates 33. Workflows to implement statistical methods that correct for diversity biases estimates now need an implementation for biologists.

5.1 Footprint of research activities

Pleiade's policy is to rely on shared computing platforms for computations that consume significant energy, for two reasons. First, those platforms have greater leverage over total energy consumption, and have the technical means to implement green computing on a useful scale. Second, those platforms have staff with the requisite skills to develop and implement policies as part of their service offer. Our partner platforms are mesocenters in Bordeaux and Grenoble, and national centers including the Idris and the CEA. Some of them charge back the cost of CO${}_2$ generation. The scientific committee of the Bordeaux mesocenter MCIA, which has a member from Pleiade, actively debates green computing implementation. Pursuant to our team policy, Pleiade does not use high-powered workstations on the desktop.

5.2 Promoting equality and diversity in science

Promoting the inclusion and diversity in science is essential at all levels: when planning science during project conception, when executing science and publishing its results, and also within the community of scientists itself. Members of Pleiade are involved in promoting the place of women in science through the participation in outreach activities, but also by committing to working groups and committees on the subject at the local and national level.

7.1 New software

7.2 New platforms

Participants: David Sherman, Ahmed Kallel.

As a founding principle, Pleiade supports reproducible scientific analyses and promotes a declarative approach using reusable software modules, rigorous documentation of data provenance, and systematic recording of workflows. The latter is a challenge when interactive interfaces are used, but can be addressed, to cite two examples, in Galaxy by extracting workflows, and in other systems by using Jupyter notebooks. Part of Pleiade's mission is to automate the deployment of environments that support these goals, for non-technical end users.

Pleiade has built a Kubernetes platform Pleiadès hosted in the Inria research center at the Univ. Bordeaux, which in 2023 was been integrated into Inria's IT Management (DSI-SP).

Use cases were identified by the project-team and from the MISTIC data management plan:

• Fast deployment of containerized user environments, combining biological data and databases, software modules specified by version, a CWL executor, and interactive tools including web front ends, notebooks, or Galaxy. A user environment will provide at least one specific HTTPS endpoint, created dynamically. A single researcher may deploy several different environments in the course of one day.
• Support for development and testing of workflows, as above but configured for team members who are developing software modules or interfaces, and who must often deploy several different environments simultaneously.
• Dynamically allocated containerized compute tasks, including both individual analysis steps in workflows and GitLab runner containers used for continuous integration. These tasks arrive in bursts that often cannot be planned in advance.
• Long-running stream preprocessing, a low-priority background task that watches external databases for changes, chooses pertinent data, precomputes representations and ingests them into local data bases.

The following requirements were derived from these use cases:

• Tasks must run in OCI containers. A typical environment will be constructed from ten to one hundred containers, grouped in Kubernetes Pods of co-localized containers that share a private network.
• Containers run unprivileged and must rely on role-based access control (RBAC), secrets, and service accounts.
• Different storage classes must be available for dynamic volume allocation: ReadWriteSingle, ReadWriteMany, Object (S3) Bucket.
• An application must be able to allocate a route with wildcard DNS in order to offer an endpoint, internally to the Inria network.
• A collection of Kubernetes custom resource definitions and RBAC definitions, specific to Pleiade's applications, is needed.
• A collection of OpenShift Operators for deployment of applications, is needed. These include database services, workflow execution, and container building using source-to-image (S2I).
• A management interface through the OKD console that allows inspection and management of app topologies, pods, volumes, and Kubernetes objects.

We support community best practices for reproducible computing in bioinformatics, using biocontainers generated by bioconda, in CWL or Galaxy workflows. For internal use we provide TES endpoints and host JupyterHub environments.

The Pleiadès platform is built on OKD 4, the community distribution of Kubernetes developed alongside of RedHat Openshift. OKD4 in particular uses the CRI-O runtime, not Docker, and containers run unprivileged. Software-defined storage and S3 endpoints are provided by Ceph. Pleiadès follows the gitops pattern and all management and implementation use Git repositories as the single source of truth.

7.3 Open data

Pleaide is strongly committed to open data and to providing machine-readable and reusable provenance with research outputs.

In the MISTIC data management plan, Pleiade and its partners specifically agreed to the following:

• Adoption of FAIR and Plan S principles.
• Data acquired by the project and provided as a deliverable will be deposited in reliable third-party archives such as data.­gouv.­fr.
• Software developed by the project will be publically hosted on gitlab.­inria.­fr during development.
• Software provided as a deliverable will be deposited in a reliable third-party archive such as inria.­hal.­science and Software Heritage, and made available under an open source license.
• Computational results provided in support of a published method or as a project deliverable will be made available in the form of a RO-Crate, providing complete, reproducible provenance.

Mutualised technical support for executing workflows with provenance is provided to Pleiade and its projects on the Pleiadès Kubernetes platform (§7.2).

8.1 Modeling metabolism of microbial communities in time and space

Participants: Pablo Ugalde-Salas, Simon Labarthe, Clémence Frioux, Coralie Muller, Rafael Augusto Kaempfer Danin, Maxime Lecomte, David Sherman.

Metabolic models allow the prediction of the metabolic behaviour of microbial strains from the knowledge of their genomes. Quantitative methods such as Flux Balance Analysis provide a framework to compute metabolic fluxes (i.e. consumption and production rates of metabolites present in the environment of the microbe) and biomass growth at the cost of an optimization problem. When modeling the dynamics of microbial communities, a FBA model must be solved for each microbial strain of the community and at each time step of the dynamics, representing an important computational load.

In 2021, an Inria Exploratory Action - SLIMMEST – carried out by Simon Labarthe and Clémence Frioux – was initiated. This project aims at combining discrete reasoning models of metabolism to numerical metamodels and PDEs for the simulation of microbial communities in time and space. The selected methodology is the coupling between PDE-based microbial population dynamics model with metamodels of complex optimizations predicting their metabolism. The main difficulty here is to ensure the scalability of the simulation and the selection of relevant metabolic functions and species to be tracked over time. The grant permitting hiring a postdoc researcher (Pablo Ugalde) and an engineer (Coralie Muller). The aim of the project is to articulate qualitative methods to analyse the functional potential of a microbial community by analysing and simplifying the community-wide metabolic capabilities, to surrogate models of the individual FBA models to provide fast and accurate simulations of the community dynamics. Qualitative methods include Answer Set Programming (ASP) to predict metabolic interactions in the community, metabolic exploration and simplifications. Surrogate models are obtained with a statistical method named ANOVA-RKHS that build a specific RKHS allowing for feature selection and promissing trade-off between speed and accuracy.

In 2023, the work achieved during the 2021 CEMRACS project was published in the CEMRACS proceedings 18. Figure 1 illustrates the use of surrogate metamodels learned by ANOVA-RKHS to speed up dynamic flux balance analysis. It was applied to a microbial community consisting of a pathogen and a commensal of the human gut microbiome. We pursued the efforts in enhancing the models by including an metabolic model of human epithelial cells interacting with the microbiota-derived chemicals diffusing in the crypt lumen. The team hosted Rafael Augusto Kaempfer Danin as intern to work on the surrogate of the epithelial cell. By increasing the speed of dynamic simulations, the latest results make possible the construction of a spatialised model of the system that would otherwise not be tractable. Figure 2 describes the biological system that is modelled and the global methodology that is followed.

The project results have been presented by Coralie Muller and Pablo Ugalde at an international conferences and an international workshop: 31st Annual Intelligent Systems For Molecular Biology and the 22nd Annual European Conference on Computational Biology (ISMB/ECCB 2023, Lyon, France – poster) 26, 7th International Symposium on Systems Biology of Microbial Infection (SBMI 2023, Jena, Germany – talk and poster) 29, 27.

The figure contains two columns. On the left is a contrast of two iterative computations: one labeled “dFBA Model” with an FBA optimization step, one labeled “Accelerated Model” with a metamodel optimization step. On the right is a flowchart showing how hyperparameters are chosen, with a trade-off between computation speed and approximation accuracy.

The upper part of the figure describes the biological system that is modelled: in absence of infection (upper left), and in infection conditions (upper right). The lower part of the figure illustrates the modelling procedure: three metabolic models are used to represent the three partners (human host, commensal bacteria and pathogen). Modelling the dynamics of the infection can be performed using dFBA or with a surrogate. Using the latter is more efficient computationally and permits building spatio-temporal models of the infection.

Another topic related to the PhD project of Maxime Lecomte consisted in integrating heterogenous omics data to build a dynamic model of cheese production. A preprint was submitted in 2023 22. It highlights a methodological approach for the construction and validation of dynamic metabolic models using dFBA. Using metabolic models for three bacterial species involved in cheese making, a careful curation and validation of metabolic functions was performed. Individual dynamics were calibrated using pure culture experimental data. The dynamics was then validated at the community level, illustrating the roles of each microbe in the production of flavour compounds.

8.2 Reasoning-based models of metabolism

Participants: Maxime Lecomte, David Sherman, Clémence Frioux, Chabname Ghassemi Nedjad, Léonard Brindel.

Over the past few years, we demonstrated how reasoning approaches can prove helpful for the analysis of microbial community metabolism through the Modeling of complementarity accross metabolic networks that permit the selection of minimal communities 45 and more generally screening metabolic potential in microbiomes 35.

These previous works illustrated the need for scalable methods in order to tackle large collections of genome, an objective that is hardly reachable with numerical models. Our recent work focuses on going further in characterizing microbial communities and in particular the interactions that occur among species and with their environment. We take advantage again of the expressivity of the logic paradigm of Answer Set Programming (ASP) as we provide a model of competitive and cooperative interactions among species in order to compare communities. As such, in the context of Maxime Lecomte's PhD, we developped scores assessing both competition and cooperation potentials based on the genome-scale metabolic networks of microbial species.

A second current axis of research relates to the PhD project of Chabname Ghassemi Nedjad. She develops an approach for the identification of seed metabolites in metabolic networks. The underlying biological challenge is to facilitate the culture of microbial dark matter, micro-organisms that cannot be cultured in controlled media experimentally. Usually, these organisms exhibit auxotrophies that prevent their growth on most nutritional conditions. We take advantage of the possibility to reconstruct metabolic networks from genomic information, therefby obtaining a blueprint of the metabolic potential of species. We apply a set of rules and constraints with the reasoning paradigm of ASP to these metabolic networks and provide sets of metabolites that would enable the activation of functions of interest in the corresponding species. Early results were presented by Chabname at a resercher summer school 23.

Our reasoning based models provide a lot of information regarding the role of each member of the microbial community, and mechanistic data on the concrete functions, genes and activity responsible for the predicted behaviour. As such, they can be very valuable. A remaining question though is the means to make all this information accessible and directly usable by the community of users who is not familiar with such modelling. To answer this question, Léonard Brindel was hired as an engineer in October 2023 and work on developing a platform associated to the sofware programme Metage2Metabo (M2M) 35. This platform will provide tools for analysing the wealth of data generated by M2M in light of additional metadata that users possess regarding their samples.

8.3 Metabolic Modeling of marine algae: methodological developments and applications

Participants: Clémence Frioux.

Brown algae, especially the species Ectocarpus siliculosus, are important models for deciphering the complex interactions within marine holobionts, with the goal of studying their metabolism together with the metabolism of the bacteria that inhabit their direct environment. Because performing wet lab experiments on such systems is technocally challenging, there is need for bioinformatic predictive methods for assessing the putative roles and interdependences between species. In parallel, addressing the difficulties brought by the study of these organisms is also a means to enhance and calibrate the tools devlopped in the team. Hence a fruitful collaboration for the past years has been developed with scientists from the Roscoff Biological station and the Inria project team Dyliss (Rennes).

The metabolism of algae is challenging to decipher, as it is for most eukaryotic species. A bottleneck resides in the difficulty of annotating their genomes, leading to manual annotations being of importance for these species. However, when aiming at comparing the metabolism of several organisms, it is preferable to annotate all of their genomes homogeneously, usually automatically, in order to prevent the propagation of biases during metabolic reconstruction. This entails ignoring the manual curation work performed on the genomes. AuCoMe is a pipeline aiming at preserving and propagating manual annotations as it permits working with heterogeneously-annotated genomes and efficiently compare their metabolic contents. The manuscript describing this pipeline was published in 2023 13. It was tested on several datasets of genomes, including algal ones, and demonstrated that it builds good quality metabolic networks as it corrects the inconsistencies brought by heterogenous annotation efforts.

8.4 Addressing the continuum of ecotypes in the gut using enterosignatures

Participants: Clémence Frioux.

The human gut microbiome generally hosts a few hundreds of microbial species, forming a dynamic but mostly stable ecosystem that can nevertheless deteriorate following perturbations. Being able to classify the state of the microbiome and identify dysbiotic conditions is of the highest interest for medical applications. Using ordination techniques or clustering approaches enables simplifying the composition of the gut microbiomes, ensuring discrete divisions and that a sample associates to one and only type 40. Such classification however lacks resolution and quantitative information regarding the proportional representation of microbial guilds in samples. To address these shortcomings, we proposed an ecologically informed decomposition of human gut microbiomes into five microbial signatures that we call 'enterosignatures' 15. The latter were obtained by applying non-negative matrix factorisation to the genus-level composition of more than 5,000 metagenomic samples.

We show that most samples of our dataset, that includes a large diversity of individuals from infants to elders, can be accurately described by a weighted combination of 2 or 3 enterosignatures. The relative proportion of each guild in the microbiome composition associates to metadata, for instance the enterosignature (ES) dominated by Firmicutes genera associates positively to health indicator, whereas we found associations between the Bacteroides-enriched ES and antibiotics consumption.

We demonstrated the generalisability of the model by applying the 5 ES to another dataset of European adults. We also assessed its relevance on an additional metacohort consisting of more than 1,100 samples from individuals with a non-western lifestyle. Results highlighted that the 5-ES model fits well to most samples, and evidenced a few cohorts with bad fitting scores whose original samples were not cold-stored, sugesting alterations in microbial community composition.

We showed that studying the samples that fit the least to the 5-ES model provides important information regarding putative perturbations of the gut microbiome. Atypical ES composition often associated to perturbed ecosystems such a antibiotic treatment or preterm birth. Altogether, we provide a simple and informative model to describe the composition of the human gut microbiome, that is generalisable to a wide range of individuals and permits highlighting perturbations of the ecosystem.

8.5 Functional profiling of the gut microbiota

Participants: Simon Labarthe.

With the emergence of metagenomic data, multiple links between the gut microbiome and the host health have been shown. Deciphering these complex interactions require evolved analysis methods focusing on the microbial ecosystem functions. Despite the fact that host or diet-derived fibres are the most abundant nutrients available in the gut, the presence of distinct functional traits regarding fibre and mucin hydrolysis, fermentation and hydrogenotrophic processes has never been investigated.

In 16, after manually selecting 91 KEGG orthologies and 33 glycoside hydrolases further aggregated in 101 functional descriptors representative of fibre and mucin degradation pathways in the gut microbiome, we used nonnegative matrix factorization to mine metagenomic datasets. Four distinct metabolic profiles were further identified on a training set of 1153 samples, thoroughly validated on a large database of 2571 unseen samples from 5 external metagenomic cohorts and confirmed with metatranscriptomic data. Profiles 1 and 2 are the main contributors to the fibre-degradation-related metagenome: they present contrasted involvement in fibre degradation and sugar metabolism and are differentially linked to dysbiosis, metabolic disease and inflammation. Profile 1 takes over Profile 2 in healthy samples, and unbalance of these profiles characterize dysbiotic samples. Furthermore, high fibre diet favours a healthy balance between Profiles 1 and Profile 2. Profile 3 takes over Profile 2 during Crohn's disease, inducing functional reorientations towards unusual metabolism such as fucose and H2S degradation or propionate, acetone and butanediol production. Profile 4 gathers under-represented functions, like methanogenesis. Two taxonomic makes up of the profiles were investigated, using either the covariation of 203 prevalent genomes or metagenomic species, both providing consistent results in line with their functional characteristics. This taxonomic characterization showed that Profiles 1 and 2 were respectively mainly composed of bacteria from the phyla Bacteroidetes and Firmicutes while Profile 3 is representative of Proteobacteria and Profile 4 of methanogens.

Integrating anaerobic microbiology knowledge with statistical learning can narrow down the metagenomic analysis to investigate functional profiles. Applying this approach to fibre degradation in the gut ended with 4 distinct functional profiles that can be easily monitored as markers of diet, dysbiosis, inflammation and disease.

8.6 Metagenomic assembly of complex microbial ecosystems

Participants: Nicolas Maurice, Léonard Brindel, Ariane Badoual, Franck Salin, Clémence Frioux.

The interest of Pleiade for the treatment of DNA sequences has renewed over the past few years with the development of a platform dedicated to the reconstruction of genomes from metagenomes, referred to as MAGs. This activity of the team is in line with the continuum of data from the sequences to the model. In 2023, Léonard Brindel worked, during his apprenticeship, on the validation of the platform initiated by Ariane Badoual.

During his intership co-supervised by Claire Lemaitre (Inria centre at the University of Rennes, Genscale team), Nicolas Maurice assessed the ability for existing metagenomic assemblers to accurately reconstruct genomes from increasingly complex communities. He developped a pipeline performing the comparison 24. This work was funded by the PEPR Agroecology and ICT. Nicolas Maurice was then hired as a PhD student in October 2023 and started a project aiming at developing new methods to facilitate the assembly of complex metagenomes such as those of the soil. He is co-supervised by Claire Lemaitre, Ricardo Vicedomini (Inria centre at the University of Rennes, Genscale team) and is hosted in the Genscale Inria team.

8.7 Statistical Modeling

Participants: Mohammed Anwar Abouabdallah, Alain Franc.

Mohamed-Anwar Abouabdallah defended his PhD thesis on Approche Tenseur-Train pour l'inférence dans les modèles à blocs stochastiques, application à la caractérisation de la biodiversité, accessible at theses.fr/2023BORD0023, on February 2nd, 2023. It presents a new approach for computing the marginals of a Weighted Stochastic Block Model from the Tensor-Train decomposition of the joint distribution. As TT decomposition provides separation of variables, computation of marginals follows swiftly. However, some numerical difficulties are met in the implementation, which have been addressed, and remain a challenging perspective. The PhD has been co-supervised by Nathalie Peurard (Inrae, MIAT, Toulouse), Olivier Coulaud (Inria BSO, Concace) and Alain Franc.

These methods are used for characterizing molecular biodiversity, building OTUs from a pairwise distance matrix using Stochastic Block Models (SBM). Building OTUs is traditionally seen as a form of unsupervised clustering. This work is done in collaboration with the MIAT INRAE research unit in Toulouse and HiePACS. It represents a connection between metabarcoding and statistical modeling, a topic which deserves investigation. Previously, in 31, we studied whether morphological-based and molecular-based approaches are in agreement, and provided evidence that yes, automatic clustering and group identification can be done reliably using barcoding. Using Aggregative Hierarchical Clustering and Stochastic Block Models, we found that the agreement between morphological-based and molecular-based classifications ranges in most cases from good to very good at taxonomic levels above species (figure 3).

The figure contains three columns comprised of vertically-oriented rectangles of different sizes and colors. Vertical intervals of the colored rectangles in adjacent columns are linked by gray horizontal paths, curved so that the paths are perpendicular to the rectangles, but change vertical position to establish the correspondence between the intervals of the rectangles. A single rectangle may thus be linked to several rectangles in the adjacent column. Overall, the intervals cover most of the extent of the rectangles. To the left, most rectangles in the middle column map to just one, or to a small number, of rectangles in the left column. To the right, the relations are more complex: rectangles in the middle column often map to several rectangles in the right column, and, conversely, rectangles in the right column may map to several rectangles in the middle column.

8.8 Metabarcoding and Taxonomic Diversity

Participants: Jean-Marc Frigerio, Alain Franc.

Metabarcoding is a series of technical procedures to build molecular based inventories from large datasets of amplicons. The underlying information needs to be compacted without losing its information content before it can be further processed with domain-specific tools. This links metabarcoding tools to dimension reduction techniques, which is an important topic in Pleiade.

Alain Franc and Jean-Marc Frigerio have continued their efforts to complete the yapotu software (§7.1.6), which makes the main stages of barcoding and metabarcoding data analysis accessible. The main progress in 2023 has been, beyond the development of some methods, the development of a Domain Specific Language, in Python, called yapsh, that makes the execution of instructions more automatic and comfortable for the user. The project, still in development, is accessible publicly on the Inria Gitlab.

The work on the evaluation of the intrinsic quality of OTUs produced in metabarcoding (Marie-Josée Cros, Jean-Marc Frigerio, Nathalie Peyrard and Alain Franc,Simple approaches for evaluation of OTU quality based on dissimilarity arrays) , has been achieved by submitting a paper to the Journal MBMG, currently in revision. The preprint of the manuscript is available at 41. As a companion to this work is available in a Git project on the public “Forge MIA” of the Inrae. It contains the documentation for the programs implementing the methods, as well as a toy data set with a tutorial on how to use it.

8.9 Dimension Reduction

Participants: Alain Franc.

Version 3 of Alain Franc's comprehensive report on Linear Dimensionality Reduction, Research Report 9488, Inria Bordeaux Sud-Ouest. 2023, 99 pages, is available at 21 and arXiv 2209.13597. The main changes from version 2 are:

1. the presentation of Quadratic embedding as an alternative to MDS when some eigenvalues of the Gram matrix are negative
2. the redaction of three appendices, on on preliminaries in linear algebra, one on quadrative forms (for quadratuic embedding), and one on a reason why random projection works.

10.1 International initiatives

10.2 International research visitors

10.3 National initiatives

10.4 Regional initiatives

11.1 Promoting scientific activities

11.2 Teaching - Supervision - Juries

11.3 Popularization

12.1 Major publications

• 1 articleP.Pedro Almeida, C.Carla Gonçalves, S.Sara Teixeira, D.Diego Libkind, M.Martin Bontrager, I.Isabelle Masneu-Pomarède, W.Warren Albertin, P.Pascal Durrens, D. J.David James Sherman, P.Philippe Marullo, C.Chris Todd Hittinger, P.Paula Gonçalves and J. P.José Paulo Sampaio. A Gondwanan imprint on global diversity and domestication of wine and cider yeast Saccharomyces uvarum..Nature Communications52014, 4044HALDOIback to text
• 2 articleR.Rodrigo Assar, M. A.Martín A Montecino, A.Alejandro Maass and D. J.David James Sherman. Modeling acclimatization by hybrid systems: Condition changes alter biological system behavior models.BioSystems121June 2014, 43-53HALDOI
• 3 articleM.Mohammad Bahram, T.Tarquin Netherway, C.Clémence Frioux, P.Pamela Ferretti, L. P.Luis Pedro Coelho, S.Stefan Geisen, P.Peer Bork and F.Falk Hildebrand. Metagenomic assessment of the global distribution of bacteria and fungi.Environmental MicrobiologyNovember 2020HALDOI
• 4 articleA.Arnaud Belcour, C.Clémence Frioux, M.Méziane Aite, A.Anthony Bretaudeau, F.Falk Hildebrand and A.Anne Siegel. Metage2Metabo, microbiota-scale metabolic complementarity for the identification of key species.eLife9December 2020HALDOIback to text
• 5 articleB.Bertille Burgunter-Delamare, H.Hetty Kleinjan, C.Clémence Frioux, E.Enora Fremy, M.Margot Wagner, E.Erwan Corre, A.Alicia Le Salver, C.Cédric Leroux, C.Catherine Leblanc, C.Catherine Boyen, A.Anne Siegel and S.Simon Dittami. Metabolic Complementarity Between a Brown Alga and Associated Cultivable Bacteria Provide Indications of Beneficial Interactions.Frontiers in Marine Science7February 2020, 1-11HALDOI
• 6 articleS. M.Simon M. Dittami, E.Erwan Corre, L.Loraine Brillet-Guéguen, A.Agnieszka Lipinska, N.Noé Pontoizeau, M.Meziane Aite, K.Komlan Avia, C.Christophe Caron, C. H.Chung Hyun Cho, J.Jonas Collen, A.Alexandre Cormier, L.Ludovic Delage, S.Sylvie Doubleau, C.Clémence Frioux, A.Angélique Gobet, I.Irene González-Navarrete, A.Agnès Groisillier, C.Cécile Herve, D.Didier Jollivet, H.Hetty Kleinjan, C.Catherine Leblanc, X.Xi Liu, D.Dominique Marie, G. .Gabriel V Markov, A. E.André E. Minoche, M.Misharl Monsoor, P.Pierre Péricard, M.-M.Marie-Mathilde Perrineau, A. F.Akira F. Peters, A.Anne Siegel, A.Amandine Siméon, C.Camille Trottier, H. S.Hwan Su Yoon, H.Heinz Himmelbauer, C.Catherine Boyen and T.Thierry Tonon. The genome of Ectocarpus subulatus – A highly stress-tolerant brown alga.Marine Genomics52January 2020, 100740HALDOI
• 7 articleC.Clémence Frioux, R.Rebecca Ansorge, E.Ezgi Özkurt, C.Chabname Ghassemi Nedjad, J.Joachim Fritscher, C.Christopher Quince, S. M.Sebastian M Waszak and F.Falk Hildebrand. Enterosignatures define common bacterial guilds in the human gut microbiome.Cell Host & MicrobeJune 2023HALDOI
• 8 articleC.Clémence Frioux, S.Simon Dittami and A.Anne Siegel. Using automated reasoning to explore the metabolism of unconventional organisms: a first step to explore host–microbial interactions.Biochemical Society Transactions483May 2020, 901-913HALDOIback to text
• 9 articleC.Clémence Frioux, D.Dipali Singh, T.Tamas Korcsmaros and F.Falk Hildebrand. From bag-of-genes to bag-of-genomes: metabolic modelling of communities in the era of metagenome-assembled genomes.Computational and Structural Biotechnology JournalJune 2020HALDOI
• 10 incollectionF.Florian Leese, A.Agnes Bouchez, K.Kessy Abarenkov, F.Florian Altermatt, A.Angel Borja, K.Kat Bruce, T.Torbjorn Ekrem, F.Fedor Ċiampor, Z.Zuzana Ċiampor, F.Filipe Costa, S.Sofia Duarte, V.Vasco Elbrecht, D.Diego Fontaneto, A. A.Alain A. Franc, M.Matthias Geiger, D.Daniel Hering, M.Maria Kahlert, B.Belma Kalamujić Stroil, M.Martyn Kelly, E.Emre Keskin, I.Igor Liska, P.Patricia Mergen, K.Kristian Meissner, J.Jan Pawlowski, L.Lyubomir Penev, Y.Yorick Reyjol, A.Ana Rotter, D.Dirk Steinke, B.Bas van der Wal, S. S.Simon S. Vitecek, J.Jonas Zimmermann and A.Alexander Weigand. Why We Need Sustainable Networks Bridging Countries, Disciplines, Cultures and Generations for Aquatic Biomonitoring 2.0: A Perspective Derived From the DNAqua-Net COST Action.Next Generation Biomonitoring: Part 158Elsevier2018, 63-99HAL
• 11 articleN. D.Nathalie Dubois Peyrard Peyrard, M.-J.Marie-Josée Cros, S.Simon De Givry, A. A.Alain A. Franc, S. S.Stephane S. Robin, R. R.Regis R. Sabbadin, T.Thomas Schiex and M.Matthieu Vignes. Exact or approximate inference in graphical models: why the choice is dictated by the treewidth, and how variable elimination can be exploited.Australian and New Zealand Journal of Statistics612to appearJune 2019, 89-133HALDOI
• 12 articleD. J.David James Sherman, T.Tiphaine Martin, M.Macha Nikolski, C.Cyril Cayla, J.-L.Jean-Luc Souciet and P.Pascal Durrens. Genolevures: protein families and synteny among complete hemiascomycetous yeast proteomes and genomes..Nucleic Acids Research372009, D550-D554HALDOIback to text

12.2 Publications of the year

12.3 Other

12.4 Cited publications

• 30 article. Génolevures – a novel approach to ‘evolutionary genomics’.FEBS Letters (complete issue)48712000, URL: https://febs.onlinelibrary.wiley.com/toc/18733468/2000/487/1back to text
• 31 articleM. A.Mohamed Anwar Abouabdallah, N.Nathalie Peyrard and A.Alain Franc. Does clustering of DNA barcodes agree with botanical classification directly at high taxonomic levels? Trees in French Guiana as a case study.Molecular Ecology Resources2022HALDOIback to textback to text
• 32 articleM.Méziane Aite, M.Marie Chevallier, C.Clémence Frioux, C.Camille Trottier, J.Jeanne Got, M. P.María Paz Cortés, S. N.Sebastián N. Mendoza, G.Grégory Carrier, O.Olivier Dameron, N.Nicolas Guillaudeux, M.Mauricio Latorre, N.Nicolás Loira, G. V.Gabriel V. Markov, A.Alejandro Maass and A.Anne Siegel. Traceability, reproducibility and wiki-exploration for “à-la-carte” reconstructions of genome-scale metabolic models.PLOS Computational Biology1452018, e1006146DOIback to text
• 33 articleB.Brian Arnold, R.Russell Corbett-Detig, D.Daniel Hartl and K.Kirsten Bomblies. RADseq underestimates diversity and introduces genealogical biases due to nonrandom haplotype sampling.Mol. Ecol.22112013, 3179--90back to text
• 34 articleM.M. Bakonyi and C. R.C. R. Johnson. The Euclidean Distance Matrix Completion Problem.SIAM J. Matrix Anal. App.1621995, 646-654back to text
• 35 articleA.Arnaud Belcour, C.Clémence Frioux, M.Méziane Aite, A.Anthony Bretaudeau, F.Falk Hildebrand and A.Anne Siegel. Metage2Metabo, microbiota-scale metabolic complementarity for the identication of key species.eLife92020, e61968DOIback to textback to text
• 36 articleM.Mélanie Bourgin, S.Simon Labarthe, A.Aicha Kriaa, M.Marie Lhomme, P.Philippe Gérard, P.Philippe Lesnik, B.Béatrice Laroche, E.Emmanuelle Maguin and M. M.Moez M. Rhimi. Exploring the Bacterial Impact on Cholesterol Cycle: A Numerical Study.Frontiers in Microbiology112020, 1121HALDOIback to text
• 37 articleE. J.E. J. Candès and B.B. Recht. Exact Matrix Completion via Convex Optimization.Found. Comput. Math.92009, 717-772back to text
• 38 bookC.Claude Combes. Parasitism: The Ecology and Evolution of Intimate Interactions.University of Chicago Press2001back to text
• 39 articleComparative genomics of protoploid Saccharomycetaceae.Genome Research192009, 1696-1709URL: http://hal.inria.fr/inria-00407511/en/DOIback to text
• 40 articleP. I.Paul I. Costea, F.Falk Hildebrand, M.Manimozhiyan Arumugam, F.Fredrik Bäckhed, M. J.Martin J. Blaser, F. D.Frederic D. Bushman, W. M.Willem M. de Vos, S. D.S. Dusko Ehrlich, C. M.Claire M. Fraser, M.Masahira Hattori, C.Curtis Huttenhower, I. B.Ian B. Jeffery, D.Dan Knights, J. D.James D. Lewis, R. E.Ruth E. Ley, H.Howard Ochman, P. W.Paul W. O’Toole, C.Christopher Quince, D. A.David A. Relman, F.Fergus Shanahan, S.Shinichi Sunagawa, J.Jun Wang, G. M.George M. Weinstock, G. D.Gary D. Wu, G.Georg Zeller, L.Liping Zhao, J.Jeroen Raes, R.Rob Knight and P.Peer Bork. Enterotypes in the landscape of gut microbial community composition.Nature Microbiology312018, 8--16DOIback to text
• 41 unpublishedM.-J.Marie-Josée Cros, J.-M.Jean-Marc Frigerio, N.Nathalie Peyrard and A.Alain Franc. Simple approaches for evaluation of OTU quality based on dissimilarity arrays.October 2022, working paper or preprintHALback to text
• 42 articleB.Bernard Dujon, D.David Sherman, G.Gilles Fischer, P.Pascal Durrens and others. Genome evolution in yeasts.Nature4302004, 35-44URL: http://www.nature.com/nature/journal/v430/n6995/abs/nature02579.htmlDOIback to text
• 43 articleO.Oliver Ebenhöh, T.Thomas Handorf and R.Reinhart Heinrich. Structural analysis of expanding metabolic networks..Genome informatics. International Conference on Genome Informatics1512004, 35--45back to text
• 44 articleA. A.Alain A. Franc, P.Pierre Blanchard and O.Olivier Coulaud. Nonlinear mapping and distance geometry.Optimization Letters142May 2019, 453-467HALDOIback to text
• 45 articleC.Clémence Frioux, E.Enora Fremy, C.Camille Trottier and A.Anne Siegel. Scalable and exhaustive screening of metabolic functions carried out by microbial consortia.Bioinformatics34172018, i934--i943DOIback to textback to textback to text
• 46 articleC.Clémence Frioux, T.Torsten Schaub, S.Sebastian Schellhorn, A.Anne Siegel and P.Philipp Wanko. Hybrid metabolic network completion.Theory and Practice of Logic Programming1912019, 83--108DOIback to text
• 47 articleP.Philippe Gayral, J.Jose Melo-Ferreira and S.Sylvain Glemin. Reference-free population genomics from next-generation transcriptome data and the vertebrate-invertebrate gap.PLoS Genetic94e10034572013back to textback to text
• 48 articleM.Martin Gebser, R.Roland Kaminski, B.Benjamin Kaufmann and T.Torsten Schaub. Clingo = ASP + Control: Preliminary Report.CoRRabs/1405.36942014back to text
• 49 inproceedingsM.Martin Gebser, R.Roland Kaminski, A.Arne Konig and T.Torsten Schaub. Advances in gringo Series 3.LPNMR6645Lecture Notes in Computer ScienceSpringer2011, 345---351back to text
• 50 articleS.Simon Labarthe, B.Bastien Polizzi, T.Thuy Phan, T.Thierry Goudon, M.Magali Ribot and B.Béatrice Laroche. A mathematical model to investigate the key drivers of the biogeography of the colon microbiota..Journal of Theoretical Biology46272019, 552-581HALDOIback to text
• 51 articleL.L. Liberti, C.C. Lavor, N.N. Maculan and A.A. Mucherino. Euclidean Distance Geometry and Applications.SIAM review56(1)2014, 3-69back to text
• 52 articleM.Michael Lynch. Estimation of Nucleotide Diversity, Disequilibrium Coefficients, and Mutation Rates from High-Coverage Genome-Sequencing Projects.Mol. Biol. Evol.25112008, 2409--19back to text
• 53 articleJ. D.Jeffrey D Orth, I.Ines Thiele and B. Ø.Bernhard Ø Palsson. What is flux balance analysis?Nature Biotechnology2832010, 245--248DOIback to text
• 54 articleS.Sylvain Prigent, C.Clémence Frioux, S. M.Simon M. Dittami, S.Sven Thiele, A.Abdelhalim Larhlimi, G.Guillaume Collet, F.Fabien Gutknecht, J.Jeanne Got, D.Damien Eveillard, J.Jérémie Bourdon, F.Frédéric Plewniak, T.Thierry Tonon and A.Anne Siegel. Meneco, a Topology-Based Gap-Filling Tool Applicable to Degraded Genome-Wide Metabolic Networks.PLOS Computational Biology1312017, e1005276DOIback to text
• 55 articleG.Guillaume Ravel, M.Michel Bergmann, A.Alain Trubuil, J.Julien Deschamps, R.Romain Briandet and S.Simon Labarthe. Inferring characteristics of bacterial swimming in biofilm matrix from time-lapse confocal laser scanning microscopy.eLife11June 2022HALDOIback to text
• 56 articleR. E.Robert E. Ricklefs. A comprehensive framework for global patterns in biodiversity.Ecology Letters712004, 1--15URL: http://dx.doi.org/10.1046/j.1461-0248.2003.00554.xDOIback to text
• 57 articleD. J.David James Sherman, P.Pascal Durrens, F.Florian Iragne, E.Emmanuelle Beyne, M.Macha Nikolski and J.-L.Jean-Luc Souciet. Genolevures complete genomes provide data and tools for comparative genomics of hemiascomycetous yeasts..Nucleic Acids Research34Database issueJanuary 2006, D432-5HALDOIback to text
• 58 articleD. J.David J. Sherman, T.Tiphaine Martin, M.Macha Nikolski, C.Cyril Cayla, J.-L.Jean-Luc Souciet and P.Pascal Durrens. Génolevures: protein families and synteny among complete hemiascomycetous yeast proteomes and genomes.Nucleic Acids Research37suppl 12009, D550-D554URL: http://nar.oxfordjournals.org/content/37/suppl_1/D550.abstractDOIback to textback to text
• 59 articleB.Benôit de Thoisy, O.Olivier Duron, L.Lo\"ic Epelboin, L.Lise Musset, P.Philippe Quénel, B.Benjamin Roche, F.Florian Binetruy, S.Sébastien Briolant, L.Luisiane Carvalho, A.Agathe Chavy, P.Pierre Couppié, M. P.Magalie Pierre Demar, M.Maylis Douine, I.I. Dusfour, S.Sébastien Gourbière, Y.Yanouk Epelboin, C.Claude Flamand, A.Alain Franc, M.Marine Ginouvès, E. S.Emeline S Houël, A.Arthur Kocher, A.Anne Lavergne, P.Paul Le Turnier, L.Luana Mathieu, J.Jérôme Murienne, M.Mathieu Nacher, R.Roxane Schaub, S.Stéphane Pelleau, G.Ghislaine Prévot, D.Dominique Rousset, E.Emmanuel Roux, S.Stanislas Talaga, P.Pauline Thill, S.Sourakhata Tirera and J.-F.Jean-François Guégan. Ecology, evolution, and epidemiology of zoonotic and vector-borne infectious diseases in French Guiana: Transdisciplinarity does matter to tackle new emerging threats.Infection, Genetics and Evolution93September 2021, 104916HALDOIback to text
• 60 articleS.Sourakhata Tirera, B.Benoit de Thoisy, D.Damien Donato, C.Christiane Bouchier, V.Vincent Lacoste, A.Alain Franc and A.Anne Lavergne. The influence of habitat on viral diversity in neotropical rodent hosts.Viruses139August 2021, 1-29HALDOIback to text