publications | PLN-team

Methodological publications

2022

The Poisson Log-Normal Model: A Generic Framework for Analyzing Joint Abundance Distributions

Julien Chiquet, Marie-Josée Cros, Mahendra Mariadassou, and 2 more authors

2022

Abs

Summary In this chapter, the authors illustrate the possibilities offered by the Poisson log-normal (PLN) model using marine species count data collected by the PISCO research program. Statistical models have been designed to respond to the limits, focusing on associations between species and their joint reaction to the environment. The presence–absence or abundance of all the species in question are also modeled jointly. The resulting models are known as joint species distribution models and are notably different from species abundance models, which consider the influence of the environment on the abundance of a single species. The authors also use the MariNet dataset, extracted from the PISCO data, to show how the PLN model may be used in response to three types of questions: evaluating the influence of covariates, such as site or year, on species abundance; identifying species which react to these covariates in the same way; and identifying direct interactions between species.

2021

Front Ecol Evol
The Poisson-Lognormal Model as a Versatile Framework for the Joint Analysis of Species Abundances

Julien Chiquet, Mahendra Mariadassou, and Stéphane Robin

Frontiers in Ecology and Evolution, 2021

Abs Bib PDF

Joint Species Distribution Models (JSDM) provide a general multivariate framework to study the joint abundances of all species from a community. JSDM account for both structuring factors (environmental characteristics or gradients, such as habitat type or nutrient availability) and potential interactions between the species (competition, mutualism, parasitism, etc.), which is instrumental in disentangling meaningful ecological interactions from mere statistical associations. Modeling the dependency between the species is challenging because of the count-valued nature of abundance data and most JSDM rely on Gaussian latent layer to encode the dependencies between species in a covariance matrix. The multivariate Poisson-lognormal (PLN) model is one such model, which can be viewed as a multivariate mixed Poisson regression model. Inferring such models raises both statistical and computational issues, many of which were solved in recent contributions using variational techniques and convex optimization tools. The PLN model turns out to be a versatile framework, within which a variety of analyses can be performed, including multivariate sample comparison, clustering of sites or samples, dimension reduction (ordination) for visualization purposes, or inferring interaction networks. This paper presents the general PLN framework and illustrates its use on a series a typical experimental datasets. All the models and methods are implemented in the R package PLNmodels, available from cran.r-project.org.
@article{PLNfrontiers, author = {Chiquet, Julien and Mariadassou, Mahendra and Robin, Stéphane}, title = {The Poisson-Lognormal Model as a Versatile Framework for the Joint Analysis of Species Abundances}, journal = {Frontiers in Ecology and Evolution}, volume = {9}, pages = {188}, year = {2021}, doi = {10.3389/fevo.2021.588292}, issn = {2296-701X}, biorxiv = {10.1101/2020.10.07.329383v1}, }

2019

ICML
Variational Inference for sparse network reconstruction from count data

Julien Chiquet, Stephane Robin, and Mahendra Mariadassou

In Proceedings of the 36th International Conference on Machine Learning, 09–15 jun 2019

Abs arXiv Bib PDF

The problem of network reconstruction from continuous data has been extensively studied and most state of the art methods rely on variants of Gaussian Graphical Models (GGM). GGM are unfortunately badly suited to sparse count data spanning several orders of magnitude. Most inference methods for count data (SparCC, REBACCA, SPIEC-EASI, gCoda, etc) first transform counts to pseudo-Gaussian observations before using GGM. We rely instead on a PoissonLogNormal (PLN) model where counts follow Poisson distributions with parameters sampled from a latent multivariate Gaussian variable, and infer the network in the latent space using a variational inference procedure. This model allows us to (i) control for confounding covariates and differences in sampling efforts and (ii) integrate data sets from different origins. It is also competitive in terms of speed and accuracy with state of the art methods.
@inproceedings{2019_ICML_CMR, title = {Variational Inference for sparse network reconstruction from count data}, author = {Chiquet, Julien and Robin, Stephane and Mariadassou, Mahendra}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {1162--1171}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, address = {Long Beach, California, USA}, month = {09--15 Jun}, publisher = {PMLR}, url = {https://proceedings.mlr.press/v97/chiquet19a.html}, }

2018

Ann Appl Stat
Variational inference for probabilistic Poisson PCA

Julien Chiquet, Mahendra Mariadassou, and Stéphane Robin

Ann. Appl. Statist., 09–15 jun 2018

Abs arXiv Bib PDF

Many application domains, such as ecology or genomics, have to deal with multivariate non-Gaussian observations. A typical example is the joint observation of the respective abundances of a set of species in a series of sites aiming to understand the covariations between these species. The Gaussian setting provides a canonical way to model such dependencies but does not apply in general. We consider here the multivariate exponential family framework for which we introduce a generic model with multivariate Gaussian latent variables. We show that approximate maximum likelihood inference can be achieved via a variational algorithm for which gradient descent easily applies. We show that this setting enables us to account for covariates and offsets. We then focus on the case of the Poisson-lognormal model in the context of community ecology. We demonstrate the efficiency of our algorithm on microbial ecology datasets. We illustrate the importance of accounting for the effects of covariates to better understand interactions between species.
@article{2018_AOAS_cmr, author = {Chiquet, Julien and Mariadassou, Mahendra and Robin, St\'ephane}, title = {Variational inference for probabilistic Poisson PCA}, journal = {Ann. Appl. Statist.}, fjournal = {The Annals of Applied Statistics}, year = {2018}, volume = {12}, number = {4}, pages = {2674-2698}, issn = {1932-6157}, doi = {10.1214/18-AOAS1177}, }

Applications of PLN in Ecology, Microbiology, Genomics

2024

ISME Comm.

A comprehensive, large-scale analysis of “terroir” cheese and milk microbiota reveals profiles strongly shaped by both geographical and human factors

Françoise Irlinger, Mahendra Mariadassou, Eric Dugat-Bony, and 13 more authors

ISME Communications, Jul 2024

2023

Anim. Microbiome

Microbiota members from body sites of dairy cows are largely shared within individual hosts throughout lactation but sharing is limited in the herd

Mahendra Mariadassou, Laurent X Nouvel, Fabienne Constant, and 13 more authors

Animal Microbiome, Jul 2023

PDF
Front. microbiol.

Microbial communities of a variety of 75 homemade fermented vegetables

Anne Thierry, Marie-Noelle Madec, Victoria Chuat, and 7 more authors

Frontiers in Microbiology, Jul 2023

2021

Ecol. Lett.
Joint species distributions reveal the combined effects of host plants, abiotic factors and species competition as drivers of species abundances in fruit flies

Benoit Facon, Abir Hafsi, Maud Charlery Masselière, and 8 more authors

Ecological Letters, Jul 2021

Abs Bib

The relative importance of ecological factors and species interactions for phytophagous insect species distributions has long been a controversial issue. Using field abundances of eight sympatric Tephritid fruit flies on 21 host plants, we inferred flies’ realized niches using joint species distribution modelling and network inference, on the community as a whole and separately on three groups of host plants. These inferences were then confronted to flies’ fundamental niches estimated through laboratory-measured fitnesses on host plants. Species abundances were mainly determined by host plants followed by climatic factors, with a minor role for competition between species sharing host plants. The relative importance of these factors mildly changed when we focused on particular host plant groups. Despite overlapping fundamental niches, specialists and generalists had almost distinct realized niches, with possible competitive exclusion of generalists by specialists on Cucurbitaceae and different assembly rules: specialists were mainly influenced by their adaptation to host plants while generalist abundances varied regardless of their fundamental host use.Competing Interest StatementThe authors have declared no competing interest.
@article{Facon2021, author = {Facon, Benoit and Hafsi, Abir and de la Masseli{\`e}re, Maud Charlery and Robin, St{\'e}phane and Massol, Fran{\c c}ois and Dubart, Maxime and Chiquet, Julien and Frago, Enric and Chiroleu, Fr{\'e}d{\'e}ric and Duyck, Pierre-Fran{\c c}ois and Ravign{\'e}, Virginie}, title = {Joint species distributions reveal the combined effects of host plants, abiotic factors and species competition as drivers of species abundances in fruit flies}, year = {2021}, doi = {10.1111/ele.13825}, biorxiv = {early/2021/03/03/2020.12.07.414326}, journal = {Ecological Letters}, }
Microbiome
Temporal patterns in Ixodes ricinus microbial communities: an insight into tick-borne microbe interactions

E Lejal, J Chiquet, J Aubert, and 12 more authors

Microbiome, Jul 2021

Abs Bib

Background Ticks transmit pathogens of medical and veterinary importance, and represent an increasing threat for human and animal health. Important steps in assessing disease risk and developing possible new future control strategies involve identifying tick-borne microbes, their temporal dynamics and interactions.Methods Using high throughput sequencing, we studied the microbiota dynamics of Ixodes ricinus from 371 nymphs collected monthly over three consecutive years in a peri-urban forest. After adjusting a Poisson Log Normal model to our data set, the implementation of a principal component analysis as well as sparse network reconstruction and differential analysis allowed us to assess inter-annual, seasonal and monthly variability of I. ricinus microbial communities as well as their interactions.Results Around 75% of the detected sequences belonged to five genera known to be maternally inherited bacteria in arthropods and potentially circulating in ticks: Candidatus Midichloria, Rickettsia, Spiroplasma, Arsenophonus and Wolbachia. The structure of the I. ricinus microbiota was temporally variable with interannual recurrence and seemed to be mainly driven by OTUs belonging to environmental genera. The total network analysis revealed a majority of positive (partial) correlations. We identified strong relationships between OTUs belonging to Wolbachia and Arsenophonus, betraying the presence of the parasitoid wasp Ixodiphagus hookeri in ticks, and the well known arthropod symbiont Spiroplasma, previously documented to be involved in the defense against parasitoid wasp in Drosophila melanogaster. Other associations were observed between the tick symbiont Candidatus Midichloria and pathogens belonging to Rickettsia, probably Rickettsia helvetica. More specific network analysis finally suggested that the presence of pathogens belonging to genera Borrelia, Anaplasma and Rickettsia might disrupt microbial interactions in I. ricinus.Conclusions Here, we identified the I. ricinus microbiota and documented for the first time the existence and recurrence of marked temporal shifts in the tick microbial community dynamics. We statistically showed strong relationships between the presence of some pathogens and the structure of the I. ricinus non-pathogenic microbes. We interestingly detected close links between some tick symbionts and the potential presence of either pathogenic Rickettsia or a parasitoid in ticks. All these new findings might be very promising for the future development of new control strategies of ticks and tick-borne diseases.Competing Interest StatementAuthor Cedric Midoux was employed by the company Irstea. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.I. ricinusIxodes ricinusPCAPrincipal Component AnalysisOTUOperational Taxonomic UnitTBPTick-Borne Pathogens
@article{Lejal2021, author = {Lejal, E and Chiquet, J and Aubert, J and Robin, S and Estrada-Pe{\~n}a, A and Rue, O and Midoux, C and Mariadassou, M and Bailly, X and Cougoul, A and Gasqui, P and Cosson, JF and Chalvet-Monfray, K and Vayssier-Taussat, M and Pollet, T}, title = {Temporal patterns in Ixodes ricinus microbial communities: an insight into tick-borne microbe interactions}, year = {2021}, volume = {9}, number = {153}, doi = {10.1186/s40168-021-01051-8}, biorxiv = {early/2020/09/26/2020.09.26.314179}, journal = {Microbiome}, }

Adv. Ecol. Res.

Chapter Nine - Coupling ecological network analysis with high-throughput sequencing-based surveys: Lessons from the next-generation biomonitoring project

Maxime Dubart, Pascal Alonso, Didac Barroso-Bergada, and 26 more authors

In The Future of Agricultural Landscapes, Part III, Jul 2021

Abs Bib

Biomonitoring ecosystems is necessary in order to evaluate risks and to efficiently manage ecosystems and their associated services. Agrosystems are the target of multiple stressors that can affect many species through effects cascading along food webs. However, classic biomonitoring, focused on species diversity or indicator species, might be a poor predictor of the risk of such whole-ecosystem perturbations. Thanks to high-throughput sequencing methods, however, it might be possible to obtain sufficient information about entire ecological communities to infer the functioning of their associated interaction networks, and thus monitor more closely the risk of the collapse of entire food webs due to external stressors. In the course of the ‘next-generation biomonitoring’ project, we collectively sought to experiment with this idea of inferring ecological networks on the basis of metabarcoding information gathered on different systems. We here give an overview of issues and preliminary results associated with this endeavour and highlight the main difficulties that such next-generation biomonitoring is still facing. Going from sampling protocols up to methods for comparing inferred networks, through biomolecular, bioinformatic, and network inference, we review all steps of the process, with a view towards generality and transferability towards other systems.

@incollection{DUBART2021367,
  title = {Chapter Nine - Coupling ecological network analysis
                    with high-throughput sequencing-based surveys:
                    Lessons from the next-generation biomonitoring
                    project},
  editor = {Bohan, David A. and Dumbrell, Alex J. and Vanbergen, Adam J.},
  series = {Advances in Ecological Research},
  publisher = {Academic Press},
  volume = {65},
  pages = {367-430},
  year = {2021},
  hal = {hal-03634351v1},
  booktitle = {The Future of Agricultural Landscapes, Part III},
  issn = {0065-2504},
  doi = {https://doi.org/10.1016/bs.aecr.2021.10.007},
  url = {https://www.sciencedirect.com/science/article/pii/S0065250421000222},
  author = {Dubart, Maxime and Alonso, Pascal and Barroso-Bergada, Didac and Becker, Nathalie and Bethune, Kevin and Bohan, David A. and Boury, Christophe and Cambon, Marine and Canard, Elsa and Chancerel, Emilie and Chiquet, Julien and David, Patrice and {de
                    Manincor}, Natasha and Donnet, Sophie and Duputié, Anne and Facon, Benoît and Guichoux, Erwan and {Le Minh}, Tâm and Ortiz-Martínez, Sebastián and Piouceau, Lucie and {Sacco-Martret de Préville}, Ambre and Plantegenest, Manuel and Poux, Céline and Ravigné, Virginie and Robin, Stéphane and Trillat, Marine and Vacher, Corinne and Vernière, Christian and Massol, François},
  keywords = {Ecological networks, Network inference,
                    Next-generation biomonitoring, High throughput
                    sequencing, Logic-based machine learning,
                    Next-generation sequencing, Microbiomes, eDNA},
}

The ISME Journal

Unraveling negative biotic interactions determining soil microbial community assembly and functioning

Sana Romdhane, Aymé Spor, Julie Aubert, and 7 more authors

The ISME Journal, Jul 2021

Preprints

2024

Composite likelihood inference for the Poisson log-normal model

Julien Stoehr, and Stephane Robin

2024

Abs arXiv

Inferring parameters of a latent variable model can be a daunting task when the conditional distribution of the latent variables given the observed ones is intractable. Variational approaches prove to be computationally efficient but, possibly, lack theoretical guarantees on the estimates, while sampling based solutions are quite the opposite. Starting from already available variational approximations, we define a first Monte Carlo EM algorithm to obtain maximum likelihood estimators, focusing on the Poisson log-normal model which provides a generic framework for the analysis of multivariate count data. We then extend this algorithm to the case of a composite likelihood in order to be able to handle higher dimensional count data.

2023

OneNet – One network to rule them all: consensus network inference from microbiome data

Camille Champion, Raphaelle Momal, Emmanuelle Le Chatelier, and 2 more authors

May 2023

2020

bioRxiv

Microbial association networks give relevant insights into plant pathobiomes

Charlie Pauvert, Tania Fort, Agnès Calonnec, and 8 more authors

May 2020

Abs

Interactions between plant pathogens and other plant-associated microorganisms regulate disease. Deciphering the networks formed by these interactions, termed pathobiomes, is crucial to disease management. Our aim was to investigate whether microbial association networks inferred from metabarcoding data give relevant insights into pathobiomes, by testing whether inferred associations contain signals of ecological interactions. We used Poisson Lognormal Models to construct microbial association networks from metabarcoding data and then investigated whether some of these associations corresponded to interactions measurable in co-cultures or known in the literature, by using grapevine (Vitis vinifera) and the fungal pathogen causing powdery mildew (Erysiphe necator) as a model system. Our model suggested that the pathogen species was associated with 23 other fungal species, forming its putative pathobiome. These associations were not known as interactions in the literature, but one of them was confirmed by our co-culture experiments. The yeast Buckleyzyma aurantiaca impeded pathogen growth and reproduction, in line with the negative association found in the microbial network. Co-cultures also supported another association involving two yeast species. Together, these findings indicate that microbial networks can provide plausible hypotheses of ecological interactions that could be used to develop microbiome-based strategies for crop protection.