Mots-Clés
systems immunology
bioinformatics
flow cytometry
immune-cell populations
machine learning
multi-omics
Behçet disease
Description
We are seeking a Master 2 student to investigate how the composition and activation states of immune-cell populations relate to depressive symptoms, fatigue and impaired quality of life in Behçet disease. This systems immunology project centres on deep immune profiling, including regulatory and effector T cells, B cells and innate immune populations. It combines analysis of flow-cytometry data with clinical, microbiome and transcriptomic measurements, using statistical methods and interpretable machine learning to identify candidate immune signatures. The project builds directly on our published studies of immune heterogeneity and systemic inflammatory diseases.
Research environment and scientific supervision
The internship will take place at the i3 laboratory (UMR-S 959, Sorbonne Université / Inserm), on the Pitié-Salpêtrière hospital campus in Paris. The laboratory studies immune regulation and inflammatory diseases, connecting biological measurements with clinical questions.
The internship will be under the scientific responsibility and direct supervision of Dr Nicolas Tchitchek, Associate Professor of Systems Immunology at Sorbonne Université. His research combines statistical analysis, interpretable machine learning and multi-omics integration to understand immune heterogeneity in inflammatory diseases. Pierre Ellul and David Saadoun will contribute expertise in psychiatry and Behçet disease.
Published findings behind the project
In our TRANSIMMUNOM study published in Annals of the Rheumatic Diseases in 2024, we analysed 443 patients across 15 disease groups and 71 healthy volunteers, measuring more than 600 immune-cell parameters with 13 flow-cytometry panels. We combined differential-abundance analyses using Wilcoxon tests and Cliff’s delta effect sizes with principal component analysis (PCA), hierarchical clustering, Spearman correlation networks and decision-tree classification. These analyses identified five disease clusters and two immunological axes associated with inflammation levels and affected tissues. The balance between LAG3 and ICOS expression in regulatory T cells contributed to the inflammatory axis, while activated Tregs and type 3 innate lymphoid cells contributed to the tissue-related axis.
In a subsequent study led by Pierre Ellul and Nicolas Tchitchek, published in Annals of the Rheumatic Diseases in 2026, we analysed 99 patients with systemic lupus erythematosus from LUPIL-2. We combined immune profiling, gut microbiome analysis and targeted metabolomics with PCA and supervised decision-tree classification to identify a signature associated with major depressive disorder, independently validated in TRANSIMMUNOM. Patients with depression showed fewer naive CD4 T cells and naive Tregs, alongside increased ICOS-positive effector-memory Tregs. These changes accompanied reduced microbial diversity and altered kynurenine and short-chain fatty acid metabolism, including lower butyrate levels. These findings raise a question for other inflammatory diseases: can comparable immune and microbial signatures be identified in Behçet disease, where depressive symptoms, fatigue and impaired quality of life contribute to the burden of illness?
Hypothesis and research objectives
We will test whether the composition and activation of immune-cell populations, together with microbial and gene-expression profiles, are associated with depressive symptoms, fatigue and quality of life in Behçet disease. By accounting for disease activity and treatment, we will assess which associations may relate more specifically to these symptoms. Comparison with lupus will help identify shared and disease-specific biological patterns.
Available data and study population
The published TRANSIMMUNOM immunophenotyping analysis included 38 patients with Behçet disease. The project will use clinical and biological datasets already curated and preprocessed in the laboratory, including deep flow cytometry, gut microbiome profiles and transcriptomics. Clinical variables include depressive symptom scores, particularly HADS-D, and, where available, fatigue, quality-of-life and anxiety scores, treatments and disease activity.
The funded project also includes an independent Behçet group of 30 patients from ImmunAID for exploratory validation of the candidate signatures. TRANSIMMUNOM will provide the discovery data, while ImmunAID will be used to examine whether the identified immune profiles are reproduced in a separate patient cohort.
Expected work
The intern will first assemble an analysis matrix linking clinical scores to available biological measurements. This will involve checking sample matching, missing data and feature distributions, then defining symptom outcomes as continuous scores and, where justified, clinically meaningful groups. Age, sex, treatments and disease activity will be considered when assessing potential confounding.
Exploratory analyses will combine descriptive statistics, principal component analysis and clustering to identify dominant sources of variation and candidate patient subgroups. Analyses will initially be conducted within each modality: immune-cell phenotypes, microbial diversity and taxonomic profiles, and gene-expression modules or pathways. Integration across modalities will focus on matched samples and biologically coherent relationships.
For signature discovery, the intern will compare parsimonious models, such as penalised regression and shallow decision trees. Random forests may provide a nonlinear comparator when supported by the sample size. Model assessment will use cross-validation, with preprocessing and feature selection performed within training folds; hyperparameter tuning will be separated from performance estimation when required. Resampling and permutation tests may assess feature stability and performance beyond chance. The objective is a reproducible, interpretable signature adapted to a small rare-disease sample.
The final phase will compare candidate signatures with our lupus findings and discuss their meaning with the clinical and immunology collaborators. This may reveal shared or disease-specific patterns involving regulatory and effector immune cells, microbial profiles or transcriptional pathways. Exploratory testing in ImmunAID will depend on clinical and biological comparability. The intern will deliver documented R or Python analyses, figures and a research report.
Expected outcomes and scientific impact
The project aims to clarify whether measurable biological profiles capture dimensions of Behçet disease that are not fully reflected by conventional disease-activity measures. Candidate signatures could guide mechanistic follow-up and larger validation studies, and inform future research on patient stratification. The expected contribution is a scientific proof of concept with transparent analytical methods; clinical utility would require further validation. The internship offers direct experience of connecting high-dimensional patient data with a defined clinical research question.
Candidate profile
We are looking for a motivated Master 2 student or final-year engineering student who is keen to work at the interface of immunology and data analysis. Curiosity about immune-cell populations, an interest in learning statistical and computational approaches, and a willingness to engage with biological interpretation are central to the project. Previous experience with R or Python is welcome.
Selected publications and methodological reading
Tchitchek N et al. Deep immunophenotyping reveals that autoimmune and autoinflammatory disorders are spread along two immunological axes capturing disease inflammation levels and types. Ann Rheum Dis. 2024.
Ellul P, Tchitchek N et al. An immunometabolic signature of major depressive disorder in systemic lupus erythematosus. Ann Rheum Dis. 2026.
Tchitchek N et al. Low-dose IL-2 shapes a tolerogenic gut microbiota that improves autoimmunity and gut inflammation. JCI Insight. 2022.
Breiman L. Random Forests. Machine Learning. 2001.
Varma S, Simon R. Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics. 2006.