Project
AA1-20
Geometric Learning for Single-Cell RNA Velocity Modeling
Project Heads
Daniel Baum, Tim Conrad, Christof Schütte, Vikram Sunkara, Christoph von Tycowicz
Project Members
Elodie Maignant
Project Duration
01.01.2024 − 31.12.2025
Located at
ZIB
Description
Recent advances in Single-Cell RNA sequencing allow to infer both the gene expression of a cell and the so-called “velocity vector” initializing the changes in that expression. In this project, we investigate how to leverage the resulting discrete vector field from a set of observations in order to recover the dynamics of the cells, that is their trajectory in the space of gene expressions. One of the challenges of this question lies in the non-linearity of the data. Indeed, the gene expression of a given type of cell has been modeled in the literature as a point of some lower-dimensional submanifold — referred as the “phenotypic manifold” — of the space of all (theoretically) possible gene expressions. Given the discrete and noisy nature of such data, it is particularly crucial to take full account of their underlying geometry to obtain correct estimates of their dynamics.
External Website
Related Publications
Sunkara, V., Rostami, A., von Tycowicz, C., & Schütte, C. (2026). Stop throwing away your Decoder; extract the learnt local coordinate system using Latent-XAI. In The 4th World Conference on Explainable Artificial Intelligence (XAI-2026).
Maignant, E., Conrad, T., & von Tycowicz, C. (2025). Tree inference with varifold distances. In International Conference on Geometric Science of Information (pp. 290-299). Cham: Springer Nature Switzerland.
Stokke, J. A., Bergmann, R., Hanik, M., & von Tycowicz, C. (2025). p-Laplacians for Manifold-valued Hypergraphs. In International Conference on Geometric Science of Information (pp. 162-171). Cham: Springer Nature Switzerland.
Related Pictures
