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. We present single-cell Embedded Topic Model (scETM). Our 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, we find 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.

show abstract

“…scETM source codes as well as the benchmarking workflows have been deposited at the GitHub repository 78…”

Section: Data Availabilitymentioning

confidence: 99%

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

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…First, the LDVAE decoder is restricted to use only linear activation functions in order to achieve interpretability; thus, LDVAE performs linear dimensionality reduction. scETM performs a similar dimensionality reduction, but further breaks down the loading matrix through tri-factorization 32 . Second, the LDVAE loss function uses a negative binomial or zero-inflated negative binomial distribution over the input features (genes), instead of the Gaussian distribution used in a classic VAE.…”

Section: Results -Sivae Accurately Generates Low Dimensional Embeddings Of Cellsmentioning

confidence: 99%

Interpretable deep generative models for genomics

Choi¹,

Quon²

2021

Preprint

View full text Add to dashboard Cite

Deep neural networks implementing generative models for dimensionality reduction have been extensively used for the visualization and analysis of genomic data. One of their key limitations is lack of interpretability: it is challenging to quantitatively identify which input features are used to construct the embedding dimensions, thus preventing insight into why cells are organized in a particular data visualization, for example. Here we present a scalable, interpretable variational autoencoder (siVAE) that is interpretable by design: it learns feature embeddings that guide the interpretation of the cell embeddings in a manner analogous to factor loadings of factor analysis. siVAE is as powerful and nearly as fast to train as the standard VAE but achieves full interpretability of the embedding dimensions. We exploit a number of connections between dimensionality reduction and gene network inference to identify gene neighborhoods and gene hubs, without the explicit need for gene network inference. Finally, we observe a systematic difference in the gene neighborhoods identified by dimensionality reduction methods and gene network inference algorithms in general, suggesting they provide complementary information about the underlying structure of the gene co-expression network.

show abstract

“…We first proposed a regularized linear decoder to include domain knowledge into autoencoders for single-cell data at a conference 86 , with scalable and expressive embeddings when compared to existing factor models, such as f-scLVM 87 . Recent approaches such as VEGA 88 , scTEM 89 and pmVAE 90 also feature VAE-based architectures with linear decoders or training separate VAEs for each GP yet connected via a global loss in the case of pmVAE. In contrast, expiMap aims toward interpretable reference mapping allowing to fuse reference atlases with GPs and enabling the query of genes or GPs.…”

Section: Discussionmentioning

confidence: 99%

Biologically informed deep learning to infer gene program activity in single cells

Lotfollahi

Rybakov

Hrovatin

et al. 2022

Preprint

View full text Add to dashboard Cite

The increasing availability of large-scale single-cell datasets has enabled the detailed description of cell states across multiple biological conditions and perturbations. In parallel, recent advances in unsupervised machine learning, particularly in transfer learning, have enabled fast and scalable mapping of these new single-cell datasets onto reference atlases. The resulting large-scale machine learning models however often have millions of parameters, rendering interpretation of the newly mapped datasets challenging. Here, we propose expiMap, a deep learning model that enables interpretable reference mapping using biologically understandable entities, such as curated sets of genes and gene programs. The key concept is the substitution of the uninterpretable nodes in an autoencoder's bottleneck by labeled nodes mapping to interpretable lists of genes, such as gene ontologies, biological pathways, or curated gene sets, for which activities are learned as constraints during reconstruction. This is enabled by the incorporation of predefined gene programs into the reference model, and at the same time allowing the model to learn de novo new programs and refine existing programs durin reference mapping. We show that the model retains similar integration performance as existing methods while providing a biologically interpretable framework for understanding cellular behavior. We demonstrate the capabilities of expiMap by applying it to 15 datasets encompassing five different tissues and species. The interpretable nature of the mapping revealed unreported associations between interferon signaling via the RIG-I/MDA5 and GPCRs pathways, with differential behavior in CD8+ T cells and CD14+ monocytes in severe COVID-19, as well as the role of annexins in the cellular communications between lymphoid and myeloid compartments for explaining patient response to the applied drugs. Finally, expiMap enabled the direct comparison of a diverse set of pancreatic beta cells from multiple studies where we observed a strong, previously unreported correlation between the unfolded protein response and asparagine N-linked glycosylation. Altogether, expiMap enables the interpretable mapping of single cell transcriptome data sets across cohorts, disease states and other perturbations.

show abstract

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

Cited by 17 publications

References 97 publications

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

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

Interpretable deep generative models for genomics

Biologically informed deep learning to infer gene program activity in single cells

Contact Info

Product

Resources

About