Hierarchical Clustering of scRNA-seq Data using TreeVAE
29.01.2025Identifying rare or novel cell populations, such as tumor cells, is essential for biomedical research. This work applies TreeVAE, a variational autoencoder tailored for hierarchical clustering of scRNA-seq data, to reveal structured patterns in gene expression and by combining this approach with a Gaussian Mixture Model we hope to be able to also find unknown cell types, e.g., tumor cells.
Understanding cellular hierarchies is crucial for studying development and disease. TreeVAE models cellular relationships in a tree-like structure, preserving biologically meaningful hierarchies. By integrating it with a Gaussian Mixture Model, we refine clustering accuracy and enhance the discovery of previously unclassified cell populations in a Semi-Unsupervised setting. This approach enables a deeper understanding of cellular diversity in scRNA-seq data, facilitating more precise classification of tumor cells and other rare populations.
Supervisor: Martin Rackl
