Interpretable generative deep learning: an illustration with single cell gene expression data

Abstract

Deep generative models can learn the underlying structure, such as pathways or gene programs, from omics data. We provide an introduction as well as an overview of such techniques, specifically illustrating their use with single-cell gene expression data. For example, the low dimensional latent representations offered by various approaches, such as variational auto-encoders, are useful to get a better understanding of the relations between observed gene expressions and experimental factors or phenotypes. Furthermore, by providing a generative model for the latent and observed variables, deep generative models can generate synthetic observations, which allow us to assess the uncertainty in the learned representations. While deep generative models are useful to learn the structure of high-dimensional omics data by efficiently capturing non-linear dependencies between genes, they are sometimes difficult to interpret due to their neural network building blocks. More precisely, to understand the relationship between learned latent variables and observed variables, e.g., gene transcript abundances and external phenotypes, is difficult. Therefore, we also illustrate current approaches that allow us to infer the relationship between learned latent variables and observed variables as well as external phenotypes. Thereby, we render deep learning approaches more interpretable. In an application with single-cell gene expression data, we demonstrate the utility of the discussed methods.

The performance of deep generative models for learning joint embeddings of single-cell multi-omics data

“…Interpretability is an aspect that is of great importance for the application of DGMs [50]. Some of the models we have reviewed already offer the possibility of making the corresponding outputs interpretable for users.…”

“…The Spearman correlations calculated between each latent variable and the features of each modality then allow relevant features to be identified. Additionally, by using a Laplace prior, scMM learns disentangled representations, with correlations between latent variables being penalized, which allows for better interpretation of individual features [55].…”

The performance of deep generative models for learning joint embeddings of single-cell multi-omics data

“…Interpretability is an aspect that is of great importance for the application of DGMs ( Treppner et al, 2022 ). Some of the models we have reviewed already offer the possibility of making the corresponding outputs interpretable for users.…”

“…The Spearman correlations calculated between each latent variable and the features of each modality then allow relevant features to be identified. Additionally, by using a Laplace prior, scMM learns disentangled representations, with correlations between latent variables being penalized, which allows for better interpretation of individual features ( Treppner et al, 2022 ).…”

Infusing structural assumptions into dimension reduction for single-cell RNA sequencing data to identify small gene sets

“…While multiple layers for dimension reduction allow for more flexibility in the learning task, which can help to construct a well-structured embedding of the data in a latent space, it is particularly difficult to determine the most explanatory genes for the learned patterns in different latent dimensions. Therefore, neural network-based approaches often rely on post-hoc feature attribution to link groups of genes to specific latent dimensions [31,32,33], where an additional analysis step is applied to an already trained model (e.g. [34,35]).…”