Scalable Single-Cell Gene Expression Generation with Latent Diffusion Models
Researchers have developed a new latent diffusion model called scLDM for generating realistic single-cell gene expression data. This model addresses challenges posed by the count nature of the data and complex gene dependencies by utilizing a permutation-invariant architecture and a diffusion transformer. The scLDM demonstrates superior performance in generating observational and perturbational single-cell data, as well as in downstream tasks like cell classification. AI
IMPACT Introduces a novel generative approach for biological data, potentially accelerating research in genomics and cellular processes.