Mots-Clés

genomics multi-omics kidney transplantation bioinformatics data integration statistics

Description

Multi-omic signature of chronic kidney allograft rejection: Integrating genetic data with the transcriptomic and clinical compartments

Supervisors: Pr. Sophie Limou (sophie.limou@univ-nantes.fr), Pr Bertrand Michel (bertrand.michel@ec-nantes.fr) and Mathieu Cap (mathieu.cap@univ-nantes.fr)
Location: Center for Research in Transplantation and Translational Immunology (CR2TI), UMR1064, CHU de Nantes, 30 bd Jean Monnet, Nantes, France. Close collaboration and mentoring from LMJL (Laboratoire de Mathématiques Jean Leray, UMR 6629, 2 Chemin de la Houssinière, Nantes, France).

Background: Kidney transplantation is currently the best treatment for end-stage kidney disease, improving both patient survival and quality of life. Development of efficient immunosuppressive treatments has significantly improved the 1-year graft survival rate, but the long-term graft survival has remained stable over the past 30 years and is still a major issue in kidney transplantation. Limiting chronic graft loss is all the more essential as the number of patients on the waiting list has been increasing for a stable number of available grafts over the last decades. Graft loss is a multifactorial event, with a central role for immune responses against the graft. However, the molecular triggers are still largely unknown and the treatment options are limited, which call for further investigation.
Preliminary results: We implemented a multi-omics project that generated multiple layers of molecular features in a well-defined clinical cohort, in order to pinpoint the molecular mechanisms at stake in chronic kidney graft rejection. We have collected transcriptomic (RNAseq for mRNAs and enhRNAs, and miRNAseq) data on 167 blood samples from kidney transplanted patients experiencing or not chronic allograft rejection. In addition, we have collected GWAS genomic data for over 4,400 individuals (over 2,000 donor-recipient pairs). Each individual molecular layer has been investigated to define molecular signatures or kidney graft rejection. We are now generating multi-omic signatures to predict graft rejection from non-invasive blood biomarkers. For that, we have leveraged several machine learning algorithms -including logistic regressions, random forests, and XGBoost- to investigate multi-omic predictive signatures. We also integrated our molecular datasets with multi-omic statistical frameworks, such as DIABLO, MOFA or RGCCA, to decompose these high-dimensional data matrices into a smaller number of interpretable latent factors involved in kidney transplant rejection. However, the current strategies cannot involve the genomic layer due to its high dimensionality (over 9 million genetic features).
Objectives: The MSc or Engineering student will explore several strategies to decrease the dimensionality of the genomic data in order to retain genetic features relevant for chronic kidney graft rejection and to enable their integration into our multi-omic frameworks. Several leads will be investigated including (i) running genome-wide association studies (GWAS) with the rejection phenotype to define a genetic risk score, (ii) limiting genetic features to functional genetic variants including coding and regulatory (eQTL) variants, (iii) assessing genetic incompatibilities from the so-called non-HLA allogenomic mismatch score, or (iv) restricting genetic features to functional variants for relevant genes or pathways from our own analyses and from the literature (e.g. BHOT panel). We will implement established statistical methods, first to identify SNPs associated with the outcome (including single-SNP logistic regression, additive genetic models, and score tests), and second to develop appropriate predictive models (including in particular Elastic Net logistic regression, partial least squares (PLS), Group LASSO, GBLUP, and genomic kernel methods). More broadly, this project will adopt a methodological perspective focused on integrating GWAS data with data from other omics and clinical compartments.
Technical skills: This internship will require the use of genomic and bioinformatic tools (e.g. Snakemake, bcftools) and the manipulation of R and Python packages for statistics and machine learning. Expertise in biostatistics and/or statistics and machine learning is required for this project.

For applications, please submit a CV and a cover letter (+ possible recommendation letters) to Sophie Limou, Bertrand Michel and Mathieu Cap.
With the financial support of the PeNSo project.