The Power of Two: integrating deep diffusion models and variational autoencoders for single-cell transcriptomics analysis

Abstract

Discovering a lower-dimensional embedding of single-cell data can greatly improve downstream analysis. The embedding should encapsulate both the high-level semantics and low-level variations in order to be meaningful and interpretable. Although current generative models have been used to learn such a low-dimensional representation, they have several limitations. Here, we introduce scVAEDer, a scalable deep-learning model that combines the power of variational autoencoders and deep diffusion models to learn a meaningful representation which can capture both global semantics and local variations in the data. By using the learned embedding, we show that scVAEDer can generate novel scRNA-seq data, predict the effect of the perturbation on various cell types, identify changes in gene expression during dedifferentiation, and detect master regulators in a biological process.