About the position
Context
Dirichlet process mixture models (DPMMs) are a Bayesian nonparametric approach to clustering in which the number of components is inferred from the data rather than prespecified 1. This makes them well suited to gene expression data, particularly single--cell R A-seq, where the number of cell types or states is usually unknown. Most existing DPMMs, however, use Gaussian components, which does not account for over-dispersed sequencing data. For mixture models in particular, the component variance plays a significant role in determining the number of components present2 , and therefore mode} misspecification can become an important issue.
ln transcriptomics studies, count-based distributions have been used for differential expression3 , normalization and variance stabilization4 as well as dimension reduction and feature selection5 On the other hand, they have received comparatively little attention as mixture kernels for nonparametric clustering. Apart from overdispersion, gene expression data sets also suffer from sparsity. Recent work shows zero-infl.ated Negative Binomial models perform better than standard multinomial or Negative Binomial models6 .
Extending the collapsed VI framework our vimixr <it package7 to such Negative Binomial kernels will likely provide both theoretical and computational edge over the state--of-theart.
Assignment
1. Antoniak, C. E. Mixtures of dirichlet processes with applications to bayesian nonparametric problems. Annals of Statistics 1152-1174 (1974) doi:10.1214/aos/1176342871.
2. Hejblum, B. P., Alkhassim, C., Gottardo, R., Caron, F. & Thiébaut, R. Sequential dirichlet process mixture of skew t-distributions for model-based clustering of fl.ow cytometry data. Annals of Applied Statistics 13, 638-660 (2019).
3. Anders, S. & Huber, vV. Differential expression analysis for sequence count data. Nature Precedings 1-1 (2010) doi:10.1186/gb-2010-11-10-r106.
4. Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome biology 20, 296 (2019).
5. Townes, F. W., Hicks, S. C., Aryee, M. J. & lrizarry, R. A. Feature selection and dimension reduction for single-cell RNA-seq based on a multinomial model. Genome biology 20, 295 (2019).
6. Wilson, T., Vo, D. H. & Thorne, T. ldentifying subpopulations of cells in single-cell transcriptomic data: A bayesian mixture modeling approach to zero inflation of counts. Journal of Computational Biology 30, 1059-1074 (2023).
7. Pal, A., Mimoun, A., Thiébaut, R. & Hejblum, B. P. Scalable dirichlet process mixture models with unknown concentration and adaptive covariance for highdimensional clustering applied to leukemia transcriptomics. arXiv Preprint (2026) doi:10.48550/arXiv.2601.21106.
Main activities
1. Develop a theoretical framework of DPMM with Poisson-Gamma/ egative Binomial mixtures, followed by a VI-based computational implementation using <it.
2. Investigate the impact of different prior choices on the model sparsity and performance.
3. Apply this clustering approach to high-dimensional transcriptomics data in the context of immunology.
Skills
- Good knowledge in Bayesian Statistics
- Programming proficiency with <it and/or O (C++ appreciated)
- An interest for exploratory analysis of biomedical data
- English proficiency (both written and spoken)
- Scientific curiosity
- Enrolled in Master 2/Engineering school with a major in (Bio )-Statistics
This listing was collected from a public source and is reproduced here for
information only. Always confirm the details on the original posting before applying.
View the original posting