Learning interpretable cellular and gene signature embeddings from single-cell transcriptomic data

The advent of single-cell RNA sequencing (scRNA-seq) technologies has revolutionized transcriptomic studies. However, large-scale integrative analysis of scRNA-seq data remains a challenge largely due to unwanted batch effects and the limited transferabilty, interpretability, and scalability of the existing computational methods.

McGill University researchers have developed single-cell Embedded Topic Model (scETM). Their key contribution is the utilization of a transferable neural-network-based encoder while having an interpretable linear decoder via a matrix tri-factorization. In particular, scETM simultaneously learns an encoder network to infer cell type mixture and a set of highly interpretable gene embeddings, topic embeddings, and batch-effect linear intercepts from multiple scRNA-seq datasets. scETM is scalable to over 10⁶ cells and confers remarkable cross-tissue and cross-species zero-shot transfer-learning performance. Using gene set enrichment analysis, the researchers found that scETM-learned topics are enriched in biologically meaningful and disease-related pathways. Lastly, scETM enables the incorporation of known gene sets into the gene embeddings, thereby directly learning the associations between pathways and topics via the topic embeddings.

scETM model overview

a scETM training. Given as input the scRNA-seq data matrices across multiple experiments or studies (i.e., batches), scETM models the single-cell transcriptomes using an embedded topic-modeling approach. Each scRNA-seq profile serves as an input to a variational autoencoder (VAE) as the normalized gene counts. The encoder network produces a stochastic sample of the latent topic mixture (θ_s,d for batch s = 1, …, S and cell d = 1, …, N_s), which can be used for clustering cells (see panel b). The linear decoder learns topic embedding and gene embedding, which can be used to analyze cellular programs via enrichment analyses (see panel c). b Workflow used to perform zero-shot transfer learning. The trained scETM-encoder on a reference scRNA-seq dataset is used to infer the cell topic mixture θ^* from an unseen scRNA-seq dataset without training them. The resulting cell mixtures are then visualized via UMAP visualization and evaluated by standard unsupervised clustering metrics using the ground-truth cell types. c Exploring gene embeddings and topic embeddings. As the genes and topics share the same embedding space, we can explore their connections via UMAP visualization or annotate each topic via enrichment analyses using known pathways.