Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data

Carilli, Maria T.; Gorin, Gennady; Choi, Yongin; Chari, Tara; Pachter, Lior

doi:10.1101/2023.01.13.523995

Cited by 10 publications

(14 citation statements)

References 91 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, mechanistic models can systematically take the experimental design into account and combine multiple types of data such as single-cell, bulk, temporal and perturbation gene expression data. Indeed, a very promising area of future work is harnessing the information contained in the multi-modal single-cell data using inference based on mechanistic models [7,19,49].…”

Section: Discussionmentioning

confidence: 99%

Bayesian model discovery for reverse-engineering biochemical networks from data

Jørgensen,

Sturrock,

Ghosh

et al. 2023

Preprint

View full text Add to dashboard Cite

The reverse engineering of gene regulatory networks based on gene expression data is a challenging inference task. A related problem in computational systems biology lies in identifying signalling networks that perform particular functions, such as adaptation. Indeed, for many research questions, there is an ongoing search for efficient inference algorithms that can identify the simplest model among a larger set of related models. To this end, in this paper, we introduce SLInG, a Bayesian sparse likelihood-free inference method using Gibbs sampling. We demonstrate that SLInG can reverse engineer stochastic gene regulatory networks from single-cell data with high accuracy, outperforming state-of-the-art correlation-based methods. Furthermore, we show that SLInG can successfully identify signalling networks that execute adaptation. Sparse hierarchical Bayesian inference thus provides a versatile tool for model discovery in systems biology and beyond.

show abstract

Section: Discussionmentioning

confidence: 99%

Bayesian model discovery for reverse-engineering biochemical networks from data

Jørgensen,

Sturrock,

Ghosh

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…We have demonstrated that mechanistic models can be used to overcome the challenge of technical noise across modalities commonly suffered by multi-view statistical methods [75, 76]. Data integration across conditions, time and modalities is an important challenge for single-cell multi-omics, including epigenome and proteome [77–79], and mechanistic models like ours could provide a versatile tool for integrating these data [80].…”

Section: Discussionmentioning

confidence: 99%

Global transcription regulation revealed from dynamical correlations in time-resolved single-cell RNA-sequencing

Volteras,

Shahrezaei,

Thomas

2023

Preprint

View full text Add to dashboard Cite

Single-cell transcriptomics reveals significant variations in the transcriptional activity across cells. Yet, it remains challenging to identify mechanisms of transcription dynamics from static snapshots. It is thus still unknown what drives global transcription dynamics in single cells. We present a stochastic model of gene expression in growing and dividing cells that harnesses temporal dimensions of single-cell RNA-sequencing through metabolic labelling protocols and cell cycle reporters. We develop a parallel and highly scalable Approximate Bayesian Computation method that corrects for technical variation and accurately quantifies absolute burst frequency, burst size and degradation rate at a transcriptome-wide scale. Using Bayesian model selection, we reveal scaling between transcription rates and cell size and unveil waves of gene regulation across the cell cycle-dependent transcriptome. Our study shows that stochastic modelling of dynamical correlations identifies global mechanisms of transcription regulation. We illustrate how mechanistic modelling and amortised inference can efficiently integrate distributions across time, modalities and conditions. Our framework thus provides a versatile tool to uncover how global transcriptional dynamics is orchestrated genome-wide in single cells.

show abstract

“…Thus, these approaches can greatly reduce the amount of time necessary to explore steady-state distributions across a wide range of transcriptional parameters for inference, parameter sensitivity investigations, and experimental design [37]. In particular, we have recently employed KWR application in a popular variational autoencoder framework to infer biophysical parameters for thousands of cells and genes, made possible because KWR, unlike generating function solutions, is compatible with gradient descent algorithms [21,27].…”

Section: Discussionmentioning

confidence: 99%

Spectral neural approximations for models of transcriptional dynamics

Gorin

Carilli

Chari

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Transcriptional systems involving discrete, stochastic events are naturally modeled using Chemical Master Equations (CMEs). These can be solved for microstate probabilities over time and state space for a better understanding of biological rates and system dynamics. However, closed form solutions to CMEs are available in only the simplest cases. Probing systems of higher complexity is challenging due to the computational cost of finding solutions and often compromises accuracy by treating infinite systems as finite. We use statistical understanding of system behavior and the generalizability of neural networks to approximate steady-state joint distribution solutions for a two-species model of the life cycle of RNA. We define a set of kernel functions using moments of the system and learn optimal weights for kernel functions with a neural network trained to minimize statistical distance between approximated and numerically calculated distributions. We show that this method of kernel weight regression (KWR) approximation is as accurate as lower-order generating-function solutions to the system, but faster; KWR approximation reduces the time for likelihood evaluation by several orders of magnitude. KWR also generalizes to produce probability predictions for system rates outside of training sets, thereby enabling efficient transcriptional parameter exploration and system analysis.

show abstract

Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data

Cited by 10 publications

References 91 publications

Bayesian model discovery for reverse-engineering biochemical networks from data

Bayesian model discovery for reverse-engineering biochemical networks from data

Global transcription regulation revealed from dynamical correlations in time-resolved single-cell RNA-sequencing

Spectral neural approximations for models of transcriptional dynamics

Contact Info

Product

Resources

About