Generating Single Cell RNA-seq Data with Diffusion Models in a Semi-Unsupervised Setting
06.10.2025Great progress has been made recently in the field of single-cell analysis. However, there are still some major hurdles, such as generating high-quality data. On the other hand, diffusion models are on everyone's lips, which would be ideal for generating new data. Furthermore, diffusion models might be able to handle biological noise better than conventional models.
Single Cell Analysis is pivotal in today's medical research, yet its effectiveness hinges on the development of robust models capable of deciphering the variability introduced by biological noise. Diffusion models, known for their ability to generate realistic samples by learning a step-by-step denoising, could therefore be adept at capturing and learning this inherent biological noise.
In this work, we want to build on an already existing method to generate single-cell data and implement methods to work with unknown samples and unknown classes.
Supervisor: Martin Rackl
