This thesis develops statistical and computational tools for the analysis of high-dimensional multivariate count data, within the framework of the Poisson Log-Normal (PLN) model and its variational inference. It introduces a Zero-Inflated PLN (ZIPLN) extension to account for excess zeros frequently observed in real count data (e.g., microbiome or single-cell data), an adaptive optimization method (AdaLVR) that improves the scalability of variational inference to datasets with several thousand variables, and corrected variational estimators for quantifying parameter uncertainty. The resulting methods are implemented in the PLNmodels and pyPLNmodels R/Python packages and applied to genomic, ecological, and microbiome count datasets.
PhD manucript
2024
PhD thesis
Machine learning for multivariate analysis of high-dimensional count data
This thesis develops statistical and computational tools for the analysis of high-dimensional multivariate count data, within the framework of the Poisson Log-Normal (PLN) model and its variational inference. It introduces a Zero-Inflated PLN (ZIPLN) extension to account for excess zeros frequently observed in real count data (e.g., microbiome or single-cell data), an adaptive optimization method (AdaLVR) that improves the scalability of variational inference to datasets with several thousand variables, and corrected variational estimators for quantifying parameter uncertainty. The resulting methods are implemented in the PLNmodels and pyPLNmodels R/Python packages and applied to genomic, ecological, and microbiome count datasets.
Journal papers
2024
arXiv
Evaluating Parameter Uncertainty in the Poisson Lognormal Model with Corrected Variational Estimators
Bastien Batardière, Julien Chiquet, and Mahendra Mariadassou
@article{batardiere2024zero,title={Zero-inflation in the Multivariate Poisson Lognormal Family},author={Batardière, Bastien and Chiquet, Julien and Gindraud, Fran{\c{c}}ois and Mariadassou, Mahendra},journal={Statistics and Computing},doi={10.1007/s11222-025-10729-0},year={2025},volume={35},number={196},}
2023
Plant Commun.
Cell specialization and coordination in Arabidopsis leaves upon pathogenic attack revealed by scRNA-seq
Etienne Delannoy, Bastien Batardière, Stéphanie Pateyron, and 4 more authors
Plant defense responses involve several biological processes that allow plants to fight against pathogenic attacks. How these different processes are orchestrated within organs and depend on specific cell types is poorly known. Here, using single-cell RNA sequencing (scRNA-seq) technology on three independent biological replicates, we identified several cell populations representing the core transcriptional responses of wild-type Arabidopsis leaves inoculated with the bacterial pathogen Pseudomonas syringae DC3000. Among these populations, we retrieved major cell types of the leaves (mesophyll, guard, epidermal, companion, and vascular S cells) with which we could associate characteristic transcriptional reprogramming and regulators, thereby specifying different cell-type responses to the pathogen. Further analyses of transcriptional dynamics, on the basis of inference of cell trajectories, indicated that the different cell types, in addition to their characteristic defense responses, can also share similar modules of gene reprogramming, uncovering a ubiquitous antagonism between immune and susceptible processes. Moreover, it appears that the defense responses of vascular S cells, epidermal cells, and mesophyll cells can evolve along two separate paths, one converging toward an identical cell fate, characterized mostly by lignification and detoxification functions. As this divergence does not correspond to the differentiation between immune and susceptible cells, we speculate that this might reflect the discrimination between cell-autonomous and non-cell-autonomous responses. Altogether our data provide an upgraded framework to describe, explore, and explain the specialization and the coordination of plant cell responses upon pathogenic challenge.
@article{delannoy2023,title={Cell specialization and coordination in Arabidopsis leaves upon pathogenic attack revealed by scRNA-seq},journal={Plant Communications},volume={4},number={5},pages={100676},year={2023},note={Focus Issue on Plant Single-Cell Biology},issn={2590-3462},doi={https://doi.org/10.1016/j.xplc.2023.100676},author={Delannoy, Etienne and Batardière, Bastien and Pateyron, Stéphanie and Soubigou-Taconnat, Ludivine and Chiquet, Julien and Colcombet, Jean and Lang, Julien},keywords={scRNA-seq, plant defense responses, plant immunity, plant susceptibility, A/ interactions, biotic stress},biorxiv={early/2023/03/02/2023.03.02.530814},}
Conferences
2024
A gradient approximation with importance sampling for dimension reduction in natural exponential families
B. Batardière, J. Chiquet, J. Kwon, and 1 more author
In actes des 55\ieme journées françaises de statistique, Bordeaux, 2024
2022
Multivariate Poisson Lognormal model: optimisation, inference and application to high dimensional data
B. Batardière, J. Chiquet, J. Kwon, and 1 more author
In actes des 53\ieme journées françaises de statistique, Lyon, 2022