Bayesian inference on stochastic gene transcription from flow cytometry data

Tiberi, Simone; Walsh, Mark D.; Cavallaro, Massimo; Hebenstreit, Daniel; Finkenstädt, Bärbel

doi:10.1093/bioinformatics/bty568

Cited by 34 publications

(33 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study, we measured gene expression in over 20,000 single cells of the fission yeast Schizosaccharomyces pombe by single molecule in situ hybridisation (smFISH, Tables S1-3 ) [21]. We combined these data with agent-based models of growing and dividing cells [22–24], stochastic models of gene expression [13,25,26] and Bayesian inference [27–33] to investigate the quantitative parameters of gene expression that mediate scaling. This integrative approach enabled us to determine which part of the transcription process is scaling with cell size and which molecular event connects transcription with the cell volume.…”

Section: Introductionmentioning

confidence: 99%

Size-dependent increase in RNA Polymerase II initiation rates mediates gene expression scaling with cell size

Sun

Bowman

Priestman

et al. 2019

Preprint

View full text Add to dashboard Cite

19Cell size varies during the cell cycle and in response to external stimuli. This requires the 20 tight coordination, or "scaling", of mRNA and protein quantities with the cell volume in order 21 to maintain biomolecules concentrations and cell density. Evidence in cell populations and 22 40Gene expression is coordinated with cell size in order to maintain biomolecule 41 concentrations. Understanding the mechanisms that mediate this coordination, called 42 hereafter "scaling", is a fundamental and intriguing problem in cell biology [1,2]. Messenger 43 RNAs (mRNA) and proteins are synthesised from the cell DNA genome, which is one of the 44 few cellular components that do not scale with size. Because cell volume increases 45 exponentially and mRNA half-lives are typically short, constant rates of mRNA or protein 46 production cannot lead to gene expression scaling. Recent work, in yeast [3], animal [4,5] 47 and plant cells [6] has shown that mRNA synthesis rates instead are coordinated globally 48 with cell size and are a major mechanism of scaling. Conversely, mRNA degradation seems 49 to be mostly unconnected to scaling [3,4,6], although evidence suggests that degradation 50 rates are adjusted early after budding yeast asymmetric division [7] and when growth rate 51 changes [8,9]. Scaling is pervasive and only few mRNAs are able to deviate from its 52 regulation [10][11][12]. Interestingly, two of them have been found to participate in the control of 53 size homeostasis [10,11]. 54What could be the molecular mechanism behind transcription scaling with cell size? For a 55 gene with an active promoter mRNA numbers follow a Poisson distribution [13]. 56Transcription is however often discontinuous and periods of RNA synthesis or 'bursts" 57 alternate with periods of promoter inactivity [14]. Work in single mammalian cells has shown 58 that scaling of mRNA numbers results from a coordination of the size of the transcription 59 bursts with cell volume and not from their frequency [4]. This is compatible with the 60 increased RNAPII occupancy observed genome-wide in large fission yeast cell cycle 61 mutants [3]. This also indicates that the mechanism behind scaling may not be related to 62 activation of transcription but rather to the efficiency of an active promoter. Critically, 63transcription is a complex process and is regulated at many levels including, RNAPII 64 initiation, pause/release, elongation, and termination [15][16][17][18]. Which of these processes is 65 coordinated with cell size to mediate scaling remains unclear. 66How could a complex set of molecular reactions such as transcription become more efficient 67 as cell size increases? In an elegant experiment Padovan-Merhar and colleagues fused cells 68 of small and large size and found that the number of mRNAs from a gene encoded in the 69 small cell genome would increase in response to the large cell environment, however its 70 concentration was almost halved. This suggests that scaling responds to both cell volume 71 and DNA content [4]. This is ...

show abstract

Section: Introductionmentioning

confidence: 99%

Size-dependent increase in RNA Polymerase II initiation rates mediates gene expression scaling with cell size

Sun

Bowman

Priestman

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…. , K. We use a Bayesian hierarchical model [21,22], which represents a natural approach to gather information from distinct samples, while allowing for sample-specific parameters, in a statistically rigorous way. We assume that X (i) was generated from a multinomial distribution:…”

Section: Resultsmentioning

confidence: 99%

BANDITS: Bayesian differential splicing accounting for sample-to-sample variability and mapping uncertainty

Tiberi

Robinson

2019

Preprint

Self Cite

View full text Add to dashboard Cite

Alternative splicing is a biological process during gene expression that allows a single gene to code for multiple proteins. However, splicing patterns can be altered in some conditions or diseases. Here, we present BANDITS, a R/Bioconductor package to perform differential splicing, at both gene and transcript-level, based on RNA-seq data.BANDITS uses a Bayesian hierarchical structure to explicitly model the variability between samples, and treats the transcript allocation of reads as latent variables. We perform an extensive benchmark across both simulated and experimental RNA-seq datasets, where BANDITS has extremely favorable performance with respect to the competitors considered.Alternative splicing plays a fundamental role in the biodiversity of proteins as it allows a single gene to generate several transcripts and, hence, to code for multiple proteins [1]. However, variations in splicing patterns can be involved in development and disregulated in disease [2][3][4].Differential splicing (DS) studies how splicing patterns vary between experimental conditions, and specifically, differential transcript usage (DTU) represents a primary branch to investigate DS [5]. DTU is present when there are changes, between two or more conditions, in the relative abundances of transcripts (i.e., in the transcript proportions), irrespective of the overall output of transcription. Alternative approaches to investigate DS are differential exon usage (DEU) [6], event specific differential splicing based on percent-spliced-in [7-9], and differential transcript expression (DTE) [5], which focuses on changes in the overall abundance of isoforms and, hence, identifies both differential gene expression (DGE) as well as differential splicing. Note that, although we broadly refer to differential splicing, all these approaches target differences in annotated transcripts (or exons), which may arise due to differential splicing as well as alternative start and terminal sites of the same transcript.A significant challenge of DTU, and in general of DS, is that transcript-level counts (i.e., the number of RNA-seq reads originating from each isoform), which are of primary interest, are not observed because most reads map to multiple transcripts (and sometimes, multiple genes).

show abstract

“…Models of transcription kinetics distinguish between two discrete promoter states: active and inactive. Importantly, these models often consider transcription as occurring in both of these states, but with large differences in rates of transcription between them [20,21,26,27]. This implies that genes which are regulated to be inactive in a tissue or cell type will tend to produce expression levels much lower than genes regulated to be active in that tissue type, resulting in a bimodal distribution for the tissue transcriptome as a whole.…”

Section: Discussionmentioning

confidence: 99%

“…Simply put, a read count of zero for a given gene does not necessarily indicate that it is inactive, while a non-zero read count does not necessarily indicate that it is actively expressed. Random transcriptional noise can often yield reads that map to inactive genes; conversely, insufficient sampling (low sequencing depth) can cause active genes to be absent among mapped reads [17][18][19][20][21][22][23][24][25][26][27]. This produces an intrinsic ambiguity between background noise and active but low-abundance genes.…”

Section: Introductionmentioning

confidence: 99%

A Hierarchical Bayesian Mixture Model for Inferring the Expression State of Genes in Transcriptomes

Thompson¹,

May²,

Moore³

et al. 2019

Preprint

View full text Add to dashboard Cite

Transcriptomes are key to understanding the relationship between genotype and phenotype. The ability to infer the expression state (active or inactive) of genes in the transcriptome offers unique benefits for addressing this issue. For example, qualitative changes in gene expression may underly the origin of novel phenotypes, and expression states are readily comparable between tissues and species. However, inferring the expression state of genes is a surprisingly difficult problem, owing to the complex biological and technical processes that give rise to observed transcriptomic datasets. Here, we develop a hierarchical Bayesian mixture model that describes this complex process, and allows us to infer expression state of genes from replicate transcriptomic libraries. We explore the statistical behavior of this method with analyses of simulated datasets-where we demonstrate its ability to correctly infer true (known) expression states-and empirical-benchmark datasets, where we demonstrate that the expression states inferred from RNA-seq datasets using our method are consistent with those based on independent evidence. The power of our method to correctly infer expression states is generally high and, remarkably, approaches the maximum possible power for this inference problem. We present an empirical analysis of primate-brain transcriptomes, which identifies genes that have a unique expression state in humans. Our method is implemented in the freely-available R package zigzag.detecting zero transcripts of a given gene in a given tissue does 53 not necessarily indicate that it is inactive. Third, even when 54 we detect transcripts of a given gene, its measured expression 55 level is likely to vary among libraries owing to both biological 56 factors (e.g., population-level variation) and technical factors 57 (i.e., the relative abundance of a given transcript in a given 58 library depends on the total transcript number of that library). 59Therefore, the rank order in expression level of two genes in 60 one library may differ from their rank order in a second library, 61 which complicates methods that infer the expression state of 62 genes based on fixed expression-level thresholds (17, 21). 63Here, we present a hierarchical Bayesian model that de-64 scribes the biological and technical processes that generate 65 transcriptomic data that-by explicitly accommodating the 66 factors described above-allows us to infer the expression state 67 of each gene from replicate RNA-seq libraries. We present anal-68 yses of simulated datasets that validate the implementation 69 and characterize the statistical behavior of our hierarchical 70 Bayesian model. We also apply our method to several em-71 pirical datasets, and demonstrate that the expression states 72 inferred using our method are consistent with expectations 73 based on independent information, such as epigenetic marks 74 and developmental-genetic studies. Finally, we demonstrate 75 our method with an empirical analysis of primate-brain tran-76 scriptomes that identi...

show abstract

Bayesian inference on stochastic gene transcription from flow cytometry data

Cited by 34 publications

References 40 publications

Size-dependent increase in RNA Polymerase II initiation rates mediates gene expression scaling with cell size

Size-dependent increase in RNA Polymerase II initiation rates mediates gene expression scaling with cell size

BANDITS: Bayesian differential splicing accounting for sample-to-sample variability and mapping uncertainty

A Hierarchical Bayesian Mixture Model for Inferring the Expression State of Genes in Transcriptomes

Contact Info

Product

Resources

About