Open Collections

Abstract

Single-cell RNA sequencing data have allowed us to gain new insights into intracellular and intercellular processes within cells. These processes can confound one another, leading to the development of methods to separate and filter such confounding signals. However, existing methods may focus on one process only or have limiting assumptions. We develop CellUntan- gler, a deep generative model, that addresses batch effects and can embed cells into a flexible latent space composed of multiple subspaces, where each subspace captures a separate biological process. We apply CellUntangler on datasets with only cycling cells and both cycling and non-cycling cells to generate embeddings that capture the cell cycle and separate information, such as cell type or differentation trajectory, where the cell cycle’s effects are removed. We demonstrate CellUntangler’s extensibility to other signals by using it to capture and separate spatial from non-spatial signals. With CellUntangler, we can obtain latent embeddings which capture different biological signals and perform enhancement or filtering at the gene expression level for downstream analyses.