Inferring single-cell transcriptomic dynamics with structured latent gene expression dynamics

Abstract

RNA velocity provides directional information for trajectory inference from single-cell RNA-sequencing data. Traditional approaches to computing RNA velocity rely on strict assumptions about the equations describing transcription of unspliced RNA and splicing of unspliced RNA into spliced RNA. This results in issues in scenarios where these assumptions are violated, such as multiple lineages with distinct dynamics and time-dependent rates. In this work we present ""LatentVelo"", a novel approach to computing a low-dimensional representation of RNA velocity with deep learning. Our approach embeds cells into a latent space with a variational auto-encoder, and describes differentiation dynamics on this latent space with neural ordinary differential equations. These more general dynamics enable accurate trajectory inference, and the latent space approach enables the generation of dynamics-based latent embeddings of cell states and batch correction of cell states and of RNA velocity. By performing simultaneous batch correction of RNA velocity dynamics and gene expression, we also outperform traditional batch correction methods that only consider gene expression. The flexible structure of the model enables modelling a variety of regulatory structures and multi-omic data, or incorporating additional information such as cell-type annotations or experimental metadata to improve the latent embedding. LatentVelo will be available at https://github.com/Spencerfar/LatentVelo.

Early human fetal lung atlas reveals the temporal dynamics of epithelial cell plasticity

“…5B and C). These identified terminal budtip cells are all at later GWs (GW [16][17][18]. Using slingshot, we identified a bifurcation in the development at early budtip cells towards either the late budtip cells, or Early AT2-like cells and NKX2-1+SOX9+CFTR+ cells (Extended Data Fig.…”

“…To determine the developmental relationships between these cells, we used our RNA velocity-based method LatentVelo 10 ( Suppl. Note Fig.…”

“…Given the large number of cells sequenced from a contiguous collection of fetal tissues, our dataset reveals important dynamic cellular plasticity and new lineage trajectories not previously identified. To determine cell lineage relationships and the ancestral origins of specific cell types, we leveraged our previously established modified RNA velocity prediction tool, LatentVelo 18 , followed by partition-based graph abstraction (PAGA) visualization and Slingshot 19 to identify several novel trajectories and show the cellular transitions of the cells from different cell origins. We identified classical trajectories previously found in the immune, stromal and endothelial compartment, but we also uncovered novel unexpected trajectories especially in the developing epithelium.…”

A dynamical perspective: moving towards mechanism in single-cell transcriptomics

“…Many have been developed: firstly, velocyto [20], which frames velocity estimation as a linear regression of unspliced and spliced reads, and subsequently several methods built to address particular modelling caveats, such as steady-state assumptions [44] or modelling timescales [45,46]. More recently, a number of studies have applied machine learning, in particular deep generative modelling, to the problem of velocity inference from splicing data [19,[47][48][49][50]. The motivation for such an approach is that while single-cell datasets may have thousands of dimensions, there exist in the data useful patterns that can be represented in a more tractable lower-dimensional space.…”

Pyro-Velocity: Probabilistic RNA Velocity inference from single-cell data

“…1b, Methods). Although several non-peerreviewed RNA velocity methods [12][13][14][15][16] have been recently proposed, here we showcase the advantages and innovations of Pyro-Velocity by comparing it with the current state-of-the-art dynamical RNA velocity model proposed by Bergen et al 1 , and implemented in scVelo.…”