Generative modeling of single-cell time series with PRESCIENT enables prediction of cell trajectories with interventions

Abstract

Existing computational methods that use single-cell RNA-sequencing (scRNA-seq) for cell fate prediction do not model how cells evolve stochastically and in physical time, nor can they predict how differentiation trajectories are altered by proposed interventions. We introduce PRESCIENT (Potential eneRgy undErlying Single Cell gradIENTs), a generative modeling framework that learns an underlying differentiation landscape from time-series scRNA-seq data. We validate PRESCIENT on an experimental lineage tracing dataset, where we show that PRESCIENT is able to predict the fate biases of progenitor cells in hematopoiesis when accounting for cell proliferation, improving upon the best-performing existing method. We demonstrate how PRESCIENT can simulate trajectories for perturbed cells, recovering the expected effects of known modulators of cell fate in hematopoiesis and pancreatic β cell differentiation. PRESCIENT is able to accommodate complex perturbations of multiple genes, at different time points and from different starting cell populations, and is available at https://github.com/gifford-lab/prescient.

Single-Cell RNA Sequencing of Human Corpus Cavernosum Reveals Cellular Heterogeneity Landscapes in Erectile Dysfunction

“…Recent advances in biotechnology and computational science have transformed the data analysis of the genome and transcriptome, holding vast potential in enhancing our understanding of cell and disease biology ( 8 ). For instance, many new approaches have been designed to facilitate a complete and detailed gene expression profile, such as identification of novel cell types and associated markers, prediction of developmental trajectories, and establishment of cell-cell interaction ( 9 – 11 ). Therefore, scRNA-seq can enable the transcriptomic profiling of thousands of cells in a single experiment and may uncover related pathological processes in a wide variety of tissues and organisms.…”

Dynamic inference of cell developmental complex energy landscape from time series single-cell transcriptomic data

“…First, GraphFP models the dynamics of cell clusters (e.g., cell states and cell types) on a discrete state space. In contrast, methods, such as Waddington-OT [ 15 ], TrajectoryNet [ 16 ] and PRESCIENT [ 17 ], modelled the dynamics of individual cells with drift-diffusion equations on a continuous state space. With the dramatic increase in amount and size of scRNA-seq data, the cluster-based approaches, which work on a relatively small number of clusters that usually represent annotated cell types, warrant both scalability to large-scale scRNA-seq data and ease of biological interpretability [ 12 ].…”

“…On the other hand, GraphFP provides an alternative and model-based approach to decipher cell-cell interactions that drive cell development. In contrast, the underlying models of both Waddington-OT [ 15 ] and PRESCIENT [ 17 ] are only able to characterize cell state-transition on the static potential energy landscape driven by random noises, failing to account for cell-cell interactions. Although able to reconstruct nonlinear development landscape, TrajectoryNet was based on the neural network framework without explicit system models, thus lacking biological interpretability.…”

“…As a remarkably rich and fruitful concept, Wasserstein distance “enables a mechanism transforming the probability space into a Riemannian manifold (known as a Wasserstein manifold), so that geometric structures and partial differential equation (PDE) techniques can be established and analyzed” [ 14 ]. Amongst existing methods, Waddington-OT [ 15 ] reported landmark work that developed an unbalanced optimal transport framework to reconstruct the cell developmental landscape by inferring cell-cell probabilistic couplings based on the distributions between adjacent time points [ 15 ]; TrajectoryNet set up a dynamic optimal transport neural network framework to reconstruct the continuous normalizing flows of evolving cell populations on the continuous state space [ 16 ]; PRESCIENT modelled cell differentiation as a diffusion process over a potential energy landscape learned by the neural network framework [ 17 ]. However, the computation of optimal transport is still a bottleneck when processing large-scale data [ 13 ].…”

PerturbNet predicts single-cell responses to unseen chemical and genetic perturbations

“…The compositional perturbation autoencoder framework [16] generates single-cell data under new combinations of observed perturbations using latent space vector arithmetic. Yeo et al (2021) proposed a generative model using a diffusion process over a potential energy landscape to learn the underlying differentiation landscape from time-series scRNA-seq data and to predict cellular trajectories under perturbations [17]. Linear models were also used to estimate the impact of perturbations on high-dimensional scRNA-seq data [4] or infer gene regulatory interactions from perturbations [18].…”