siVAE: interpretable deep generative models for single-cell transcriptomes

Abstract

Neural networks such as variational autoencoders (VAE) perform dimensionality reduction for the visualization and analysis of genomic data, but are limited in their interpretability: it is unknown which data features are represented by each embedding dimension. We present siVAE, a VAE that is interpretable by design, thereby enhancing downstream analysis tasks. Through interpretation, siVAE also identifies gene modules and hubs without explicit gene network inference. We use siVAE to identify gene modules whose connectivity is associated with diverse phenotypes such as iPSC neuronal differentiation efficiency and dementia, showcasing the wide applicability of interpretable generative models for genomic data analysis.

Behind the Indolent Facade: Uncovering the Molecular Features and Malignancy Potential in Lung Minimally Invasive Adenocarcinoma by Single‐Cell Transcriptomics

“…This subcluster comprises cells from different patients, thereby excluding the potential influence of interpatient heterogeneity on the trajectory (Figure S3A, Supporting Information). Subsequently, we employed the slingshot [ 18 ] and scTour [ 19 ] method to infer the state trajectories, which yielded similar paths and developmental dynamics (Figure S3B,C, Supporting Information), affirming the robustness of the results obtained through Monocle3. Our findings highlight a distinctive cell state within MIA epithelial cells and implies a basal‐like molecular nature for a predominant subset of pre‐malignant epithelial cells in MIA.…”

“…Trajectory Inferencse and Pseudotime Analysis: Monocle3, [54] slingshot [18] and scTour [19] were used for the trajectory inference and pseudotime analysis. For Monocle3, a principal graph was created by using the "learn_graph" function and selected the root position programmatically by using a helper function provided by the author of Monocle3.…”

scBiG for representation learning of single-cell gene expression data based on bipartite graph embedding

“…For reconstruction of gene expression, we compared it with five alternative methods, including MAGIC ( 26 ), scImpute ( 27 ), ALRA ( 28 ), DCA ( 7 ) and scVI ( 9 ). For gene co-expression analysis, we compared it with siVAE ( 16 ) and SIMBA ( 15 ). Supplementary Table S2 provides a brief summary of the alternative methods and corresponding software tools.…”

“…For example, SIMBA ( 15 ) utilizes graph structures to co-embed cells and various features, such as genes and open chromatin regions, into a shared latent space and uses these embeddings for downstream analysis. siVAE ( 16 ) is a deep generative model that uses pairs of encoders and decoders to learn representations of genes and cells in the latent space based on the variational autoencoder framework.…”

BuDDI:Bulk Deconvolution with Domain Invarianceto predict cell-type-specific perturbations from bulk

“…There exist several generative methods to learn interpretable latent spaces that decompose the input single-cell expression profiles into relevant sources of variation. These methods can be directly trained to capture a specific source of variation [29][30][31][32][33][34][35] or post-hoc-interpreted after training [36][37][38][39][40] . Furthermore, there exist several methods to learn a latent space such that shifts within the latent space represent specific perturbation effects on an unobserved cell or cell type [4][5][6][7][8][9][10][11][12][13][14]28 .…”