Deep Generative Modeling for Single-cell Transcriptomics

By Romain Lopez, Jeffrey C. Regier, Michael Cole, Michael I. Jordan, Nir Yosef. Nature Methods, 2018

2,423 citations recorded by the source corpus.

Single-cell transcriptome measurements can reveal unexplored biological diversity, but they suffer from technical noise and bias that must be modeled to account for the resulting uncertainty in downstream analyses. Here we introduce single-cell variational inference (scVI), a ready-to-use scalable framework for the probabilistic representation and analysis of gene expression in single cells (https://github.com/YosefLab/scVI). scVI uses stochastic optimization and deep neural networks to aggregate information across similar cells and genes and to approximate the distributions that underlie observed expression values, while accounting for batch effects and limited sensitivity. We used scVI for a range of fundamental analysis tasks including batch correction, visualization, clustering, and differential expression, and achieved high accuracy for each task. scVI is a ready-to-use generative deep learning tool for large-scale single-cell RNA-seq data that enables raw data processing and a wide range of rapid and accurate downstream analyses.

Subfields: Biophysics, Molecular Biology

View Paper: https://www.semanticscholar.org/paper/ed5d6be2eb8fc1d42f9a0e3c50abaf5439bbd8dd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biophysics

In Molecular Biology

Citations in the corpus

MIWAE: Deep Generative Modelling and Imputation of Incomplete Data Sets (2019, background): “In recent years, DLVMs have had remarkable successes in modelling images (Siddharth et al., 2017), molecules (Kusner et al., 2017; Gómez-Bombarelli et al., 2018; Riesselman et al., 2018), arithmetic expressions (Kusner et al., 2017), and single-cell data (Grønbech et al., 2018; Lopez et al., 2018).”
Comprehensive integration of single-cell data (2018, positive): “New probabilistic approaches for scRNA-seq data normalization and analysis using neural networks have also been recently introduced, with the advantage that they scale to very large datasets and explicitly model batch effects (Lopez et al., 2018).”
The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans (2021, background): “S1 to S16 Tables S1 to S9 References (40-95)”
How to Hallucinate Functional Proteins (2019, background): “Impressive results have been published where this class of models has generated realistic examples of faces (8; 9), audio (10), 3D objects (11), and RNA expression profiles (12).”
Predicting transcriptional outcomes of novel multigene perturbations with GEARS (2023, positive): “GEARS functionalizes this graph-based inductive bias using a graph neural network (GNN) architecture 33 .”
Feature selection and dimension reduction for single-cell RNA-Seq based on a multinomial model (2019, methodology): “The variational autoencoder (VAE, a type of neural network) method scVI [47] utilizes a negative binomial likelihood in the decoder, while the encoder relies on log-normalized input data for numerical stability.”
AUTO-ENCODING VARIATIONAL BAYES (2020, neutral): “Single-cell Variational Inference (scVI, [7], Appendix G further details the generative model) models scRNA-seq data with a latent variables hng representing the underlying gene expression level for gene g in cell n, corrected for a certain number of technical variations (e.”
Spatial components of molecular tissue biology (2022, neutral): “…molecular profiles are available, cellular phenotypes can be defined based on both organ-elle properties (for example, nucleus size) and the distribution characteristics of molecules across the cell, including their covariance structure (to define complexes, membraneless compartments and so on).”
Cross-tissue immune cell analysis reveals tissue-specific features in humans (2022, methodology): “Data integration was done using BBKNN (70) and scVI (71), and the results were compared using kBET (72).”