Multi-view machine learning analysis of intestinal immune response initiation in Peyer’s patches

Host laboratory and collaborators


Cécile CAPPONI (LIS) /


Peyer’s patches are the main inductive sites of intestinal immunity, which is crucial for host defense against pathogens but also for regulation of the microbiota and tolerance to food proteins. Both right positioning and activation of specific immune cells are critical for the drastic change of transcriptional programs that lead to the initiation of the intestinal immune response. However, the links between immune cell location, their migration and interaction with neighboring cells , activation, transcriptional changes, and the immune response itself remain poorly characterized. Here, we will combine spatial transcriptomics, high resolution fluorescent imaging, bioinformatic analysis and cross-modal machine learning methods to decipher the microenvironmental circuits involved in Peyer’s patch immune response initiation in different conditions of stimulation. It will allow to understand how these different cellular mechanisms are orchestrated together to produce an appropriate immune response to detected threats.


Intestinal immunity ; Spatial transcriptomics ; High-resolution fluorescent imaging ; Cross-modal machine learning


1. Determine spatial organization, interactions and transcriptome of immune cells within Peyer’s patches at steady state and upon different conditions of stimulation.
2. Develop innovative bioinformatic methods to analyze in a unify mode multi-modal data.
3. Study the required properties of learning algorithms for fitting the biological problem and design a relevant cross-modal learning algorithm with both theoretical and experimental guarantees.
4. Infer spatially and timely-defined factors and signals likely to drive immune response initiation.

Proposed approach (experimental / theoretical / computational)

The experimental approach consists in the acquisition of biological data mainly obtained by spatial transcriptomics coupled to high-resolution immunofluorescence imaging. This will allow to obtain in each transcriptional spot cell identities, cell numbers and cell-cell interactions in the different conditions of stimulation. The computational aspect is the bioinformatic analysis applied to these data. Analysis of spatial transcriptomics, a very dynamic field in constant development, will require application of the best existing methods coupled with custom development to tackle our specific biological questions. In particular, development of cross-modal machine learning methods will permit to produce innovative analysis and prediction methods. The recruited learning methods will be studied through their ability to guarantee «better than baseline» prediction accuracy, under the prior and computed hypothesis on the various views of data description, independently of each other, but also over their potential correlations.


The project aims to (i) develop experimental methods to capture high quality spatial fluorescent images with a process compatible with single-cell transcriptomic capture; (ii) improve existing methods in bioinformatics to analyse spatial transcriptomic information at several levels (gene expression, cell-cell contact, cell identities, stimulation conditions and kinetics,...); (iii) discover and formalise new settings of multi-view machine learning tasks, and solid methods to address them. This project thus gathers 3 scientific fields : biology, bioinformatics and artificial intelligence.

Expected profile

Good knowledge of single-cell RNA-seq analysis methods, differential expression calculations and functional enrichment. Basic knowledge of imaging analysis. Interest in biological questions, in particular in immunology and cell biology. Good organisational skills with synthetic presentation of scientific result and good communication with mathematicians, bioinformaticians and biologists.

Is this project the continuation of an existing project or an entirely new one? In the case of an existing project, please explain the links between the two projects

It is a new project based on the skills of supervisor 1 in Peyer’s patch biology and of supervisor2 in multimodal machine learning.

2 to 5 references related to the project

1- Ståhl PL, Salmén F, Vickovic S, Lundmark A, Navarro JF, Magnusson J, Giacomello S, Asp M, Westholm JO, Huss M, Mollbrink A, Linnarsson S, Codeluppi S, Borg Å, Pontén F, Costea PI, Sahlén P, Mulder J, Bergmann O, Lundeberg J, Frisén J. Visualization and analysis of gene expression in tissue sections by spatial transcriptomics. Science. 2016 Jul 1;353(6294):78-82.
2- Wagner, C., Bonnardel, J., Da Silva, C., Spinelli, L., Arroyo Portilla, C., Tomas, J., Lagier, M., Chasson, L., Masse, M., Dalod, M., Chollat-Namy, A., Gorvel J.P. and Lelouard, H.(2020) Differentiation paths of Peyer’s patch LysoDC are linked to sampling site positioning, migration and T cell priming. Cell reports. 31, 107479.
3- Guilliams M., Bonnardel J. , Haest B., Vanderborght B. et al. (2021) Spatial proteogenomics reveals distinct and evolutionarily-conserved hepatic macrophage niches. bioRxiv. 2021.10.15.464432
4- T. Baltrušaitis, C. Ahuja and L. Morency (2019), "Multimodal Machine Learning: A Survey and Taxonomy," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 41, no. 2, pp. 423-443, doi: 10.1109/TPAMI.2018.2798607.
5- Riikka Huusari, Hachem Kadri, Cécile Capponi (2018), « Multi-view Metric Learning in Vector-valued Kernel Spaces » Proceedings of the Twenty-First International Conference on Artificial Intelligence and Statistics, PMLR 84:415-424.

3 main publications from each PI over the last 5 years

H. Lelouard:
1- Luciani C, Hager FT, Cerovic V, Lelouard H. (2021) Dendritic cell functions in the inductive and effector sites of intestinal immunity. Mucosal Immunol. doi: 10.1038/s41385-021-00448-w.
2- Wagner, C., Bonnardel, J., Da Silva, C., Spinelli, L., Arroyo Portilla, C., Tomas, J., Lagier, M., Chasson, L., Masse, M., Dalod, M., Chollat-Namy, A., Gorvel J.P. and Lelouard, H.(2020) Differentiation paths of Peyer’s patch LysoDC are linked to sampling site positioning, migration and T cell priming. Cell reports. 31, 107479.
3-Bonnardel, J., Da Silva, C., Wagner, C., Bonifay, R., Chasson, L., Masse, M., Pollet, E., Dalod, M., Gorvel, J. P., and Lelouard, H. (2017) Distribution, location, and transcriptional profile of Peyer's patch conventional DC subsets at steady state and under TLR7 ligand stimulation. Mucosal Immunol. 10, 1412-1430.

C. Capponi:
1- Q Ferré, J Chèneby, D Puthier, C Capponi, B Ballester (2021), « Anomaly detection in genomic catalogues using unsupervised multi-view autoencoders », BMC bioinformatics 22 (1), 1-26.
2- B Bauvin, C Capponi, JF Roy, F Laviolette (2020), « Fast greedy C -bound minimization with guarantees », Machine Learning 109 (9), 1945-1986.
3-Dominique Benielli, Cécile Capponi, Baptiste Bauvin, Sokol Koço, Hachem Kadri, Riikka Huusari, François Laviolette (2021), Toolbox for Multimodal Learn (scikit-multimodallearn), hal-03473134, to appear in JMLR Software, 2022.