To the Editor -Methods for analyzing single-cell data 1-4 perform a core set of computational tasks. These tasks include dimensionality reduction, cell clustering, cell-state annotation, removal of unwanted variation, analysis of differential expression, identification of spatial patterns of gene expression, and joint analysis of multi-modal omics data. Many of these methods rely on likelihood-based models to represent variation in the data; we refer to these as 'probabilistic

RNA velocity has been rapidly adopted to guide the interpretation of transcriptional dynamics in snapshot single-cell transcriptomics data. Current approaches for estimating and analyzing RNA velocity can empirically reveal complex dynamics but lack effective strategies for quantifying the uncertainty of the estimate and its overall applicability to the system of interest. Here, we present veloVI (velocity variational inference), a deep generative modeling framework for estimating RNA velocity. veloVI learns a gene-specific dynamical model of RNA metabolism and provides a transcriptome-wide quantification of velocity uncertainty. We show in a series of examples that veloVI compares favorably to previous approaches for inferring RNA velocity with improvements in fit to the data, consistency across transcriptionally similar cells, and stability across preprocessing pipelines for quantifying RNA abundance. Further, we demonstrate that properties unique to veloVI, such as posterior velocity uncertainty, can be used to assess the appropriateness of analysis with velocity to the data at hand. Finally, we highlight veloVI as a flexible framework for modeling transcriptional dynamics by adapting the underlying dynamical model to use time-dependent transcription rates.

No abstract

Direct comparison of bulk gene expression profiles is complicated by distinct cell type mixtures in each sample that obscure whether observed differences are actually caused by changes in the expression levels themselves or are simply a result of differing cell type compositions. Single-cell technology has made it possible to measure gene expression in individual cells, achieving higher resolution at the expense of increased noise. If carefully incorporated, such single-cell data can be used to deconvolve bulk samples to yield accurate estimates of the true cell type proportions, thus enabling one to disentangle the effects of differential expression and cell type mixtures. Here, we propose a generative model and a likelihood-based inference method that uses asymptotic statistical theory and a novel optimization procedure to perform deconvolution of bulk RNA-seq data to produce accurate cell type proportion estimates. We show the effectiveness of our method, called RNA-Sieve, across a diverse array of scenarios involving real data and discuss extensions made uniquely possible by our probabilistic framework, including a demonstration of well-calibrated confidence intervals.

Contemporary single-cell omics technologies have enabled complex experimental designs incorporating hundreds of samples accompanied by detailed information on sample-level conditions. Current approaches for analyzing condition-level heterogeneity in these experiments often rely on a simplification of the data such as an aggregation at the cell-type or cell-state-neighborhood level. Here we present MrVI, a deep generative model that provides sample-sample comparisons at a single-cell resolution, permitting the discovery of subtle sample-specific effects across cell populations. Additionally, the output of MrVI can be used to quantify the association between sample-level metadata and cell state variation. We benchmarked MrVI against conventional meta-analysis procedures on two synthetic datasets and one real dataset with a well-controlled experimental structure. This work introduces a novel approach to understanding sample-level heterogeneity while leveraging the full resolution of single-cell sequencing data.

Direct comparison of bulk gene expression profiles is complicated by distinct cell type mixtures in each sample which obscure whether observed differences are actually due to changes in expression levels themselves or simply cell type compositions. Single-cell technology has made it possible to measure gene expression in individual cells, achieving higher resolution at the expense of increased noise. If carefully incorporated, such single-cell data can be used to deconvolve bulk samples to yield accurate estimates of the true cell type proportions, thus enabling one to disentangle the effects of differential expression and cell type mixtures. Here, we propose a generative model and a likelihood-based inference method that uses asymptotic statistical theory and a novel optimization procedure to perform deconvolution of bulk RNA-seq data to produce accurate cell type proportion estimates. We demonstrate the effectiveness of our method, called RNA-Sieve, across a diverse array of scenarios involving real data and discuss several extensions made uniquely possible by our probabilistic framework, including general hypotheses tests and confidence intervals.