<i>DeepVelo</i>
            : Single-cell transcriptomic deep velocity field learning with neural ordinary differential equations

Chen, Zhanlin; King, William C.; Hwang, Ahyeon; Gerstein, Mark; Zhang, Jing

doi:10.1126/sciadv.abq3745

Cited by 44 publications

(71 citation statements)

References 81 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We additionally attempted to use a lower dimensional embedding of the data as input to the neural ODE to determine if this would help with the problem of noise in inherently stochastic data. A variational autoencoder (VAE) has been used with neural network models in other genomics analysis methods for this purpose (Lotfollahi et al, 2019), (Yu and Welch, 2022), (Chen et al, 2022). To test this approach, we trained a VAE on the simulated data from the first two time points.…”

Section: Rnaforecaster Makes Accurate Predictions In Future Expressio...mentioning

confidence: 99%

“…However, these RNA velocity and protein acceleration methods do not make predictions about future or past cell states beyond the immediate changes in expression. To predict expression state changes further into the future, vector fields have been applied to the concept of RNA velocity to allow for prediction of future states (Qiu et al, 2022), (Chen et al, 2022). One of these methods, called Dynamo, additionally suggests the use of metabolic labeling scRNAseq variants (Battich et al, 2020), (Qiu et al, 2020), (Hendriks et al, 2019), (Erhard et al, 2019), (Cao et al, 2020), in which cells are treated with a modified uridine for a set period of time before they are harvested for sequencing.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Transcriptomic forecasting with neural ODEs

Erbe

Stein-O’Brien

Fertig

2022

Preprint

View full text Add to dashboard Cite

Single-cell RNA-seq (scRNA-seq) technologies can uncover changes in the molecular states that underlie cellular phenotypes. However, to understand dynamic cellular processes, computational tools are needed to extract temporal information from the snapshots of cellular gene expression that scRNA-seq provides. To address this challenge, we have developed a neural ordinary differential equation based method, RNAForecaster, for predicting gene expression states in single cells for multiple future time steps. We demonstrate that RNAForecaster can accurately predict future expression states in simulated scRNA-seq data. We then show that using metabolic labeling scRNA-seq data from constitutively dividing cells, RNAForecaster accurately recapitulates many of the expected changes in gene expression during progression through the cell cycle over a three day period. Finally, we extend RNAForecaster to use unspliced and spliced counts from scRNA-seq to predict the impact of knock down experiments in pancreatic development. Thus, RNAForecaster enables estimation of future expression states in biological systems.

show abstract

Section: Rnaforecaster Makes Accurate Predictions In Future Expressio...mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Transcriptomic forecasting with neural ODEs

Erbe

Stein-O’Brien

Fertig

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Moreover, it may often be important to separate the counts of molecules of different splicing status, allocating the associated UMIs as having a spliced or unspliced (or sometimes ambiguous) origin. For example, analyses based on splicing status-aware counts have been used to study (and simulate) cell differentiation and development processes (4)(5)(6)(7)(8)(9) and have proven useful in other important research endeavors, such as disease state prediction (10). Therefore, it is important to understand, analyze, and improve the methods by which the splicing status of UMIs is inferred.…”

Section: Introductionmentioning

confidence: 99%

Understanding and evaluating ambiguity in single-cell and single-nucleus RNA-sequencing

Soneson

Patro

2023

Preprint

View full text Add to dashboard Cite

Recently, a new modification has been proposed by Hjörleifsson and Sullivan et al. to the model used to classify the splicing status of reads (as spliced (mature), unspliced (nascent), or ambiguous) in single-cell and single-nucleus RNA-seq data. Here, we evaluate both the theoretical basis and practical implementation of the proposed method. The proposed method is highly-conservative, and therefore, unlikely to mischaracterize reads as spliced (mature) or unspliced (nascent) when they are not. However, we find that it leaves a large fraction of reads classified as ambiguous, and, in practice, allocates these ambiguous reads in an all-or-nothing manner, and differently between single-cell and single-nucleus RNA-seq data. Further, as implemented in practice, the ambiguous classification is implicit and based on the index against which the reads are mapped, which leads to several drawbacks compared to methods that consider both spliced (mature) and unspliced (nascent) mapping targets simultaneously — for example, the ability to use confidently assigned reads to rescue ambiguous reads based on shared UMIs and gene targets. Nonetheless, we show that these conservative assignment rules can be obtained directly in existing approaches simply by altering the set of targets that are indexed. To this end, we introduce the spliceu reference and show that its use with alevin-fry recapitulates the more conservative proposed classification. We also observe that, on experimental data, and under the proposed allocation rules for ambiguous UMIs, the difference between the proposed classification scheme and existing conventions appears much smaller than previously reported. We demonstrate the use of the new piscem index for mapping simultaneously against spliced (mature) and unspliced (nascent) targets, allowing classification against the full nascent and mature transcriptome in human or mouse in <3GB of memory. Finally, we discuss the potential of incorporating probabilistic evidence into the inference of splicing status, and suggest that it may provide benefits beyond what can be obtained from discrete classification of UMIs as splicing-ambiguous.

show abstract

“…Cellular dynamics is stochastic and can be genetically modeled as, e.g., a convection-diffusion process (1, 8). Dynamo and a subsequently developed DeepVelo approach (9) reconstruct the convection part only.…”

Section: Introductionmentioning

confidence: 99%

Graph-Dynamo: Learning stochastic cellular state transition dynamics from single cell data

Zhang,

Qiu,

et al. 2023

Preprint

View full text Add to dashboard Cite

Modeling cellular processes in the framework of dynamical systems theories is a focused area in systems and mathematical biology, but a bottleneck to extend the efforts to genome-wide modeling is lack of quantitative data to constrain model parameters. With advances of single cell techniques, learning dynamical information from high throughput snapshot single cell data emerges as an exciting direction in single cell studies. Our previously developed dynamo framework reconstructs generally nonlinear genome-wide gene regulation relations from single cell expression state and either splicing- or metabolic labeling-based RNA velocity data. In this work, we first developed a graph-based machine learning procedure that imposes a mathematical constraint that the RNA velocity vectors lie in the tangent space of the low-dimensional manifold formed by the single cell expression data. Unlike a popular cosine correlation kernel used in literature, this tangent space projection (TSP) preserves the magnitude information of a vector when one transforms between different representations of the data manifold. Next, we formulated a data-driven graph Fokker-Planck (FPE) equation formalism that models the full cellular state transition dynamics as a convection-diffusion process on a data-formed graph network. The formalism is invariant under representation transformation and preserves the topological and dynamical properties of the system dynamics. Numerical tests on synthetic data and experimental scRNA-seq data demonstrate that the graph TSP/FPE formalism built from snapshot single cell data can recapitulate system dynamics.

show abstract

DeepVelo : Single-cell transcriptomic deep velocity field learning with neural ordinary differential equations

Cited by 44 publications

References 81 publications

Transcriptomic forecasting with neural ODEs

Transcriptomic forecasting with neural ODEs

Understanding and evaluating ambiguity in single-cell and single-nucleus RNA-sequencing

Graph-Dynamo: Learning stochastic cellular state transition dynamics from single cell data

Contact Info

Product

Resources

About