ANC Workshop - 28/04/2020



Douglas Armstrong


Christos Maniatis


Bayesian modelling for single cell multi-omics data


Significant advancements in next-generation sequencing (NGS) have revolutionized the way we map the epigenetic landscape. Traditional NGS approaches provide molecular measurements at a bulk level where gene expression and epigenomic features are measured as an average across thousands or even millions of cells. Recently, single cell multi-omics technologies like single cell Methylome and Transcriptome sequencing (sc-MT) and single cell Nucleosome, Methylation and Transcription sequencing (sc-NMT) arose as promising tools to eliminate confounding phenomena by capturing data at a single cell resolution. Despite the potential of these technologies, low coverage, data sparsity and lack of noiseaware practices prevent regulatory relationships between different biological layers to emerge. In this work, we present a Bayesian hierarchical model in which correlation between different multi-omics measurements is captured through a latent multivariate Gaussian distribution. Furthermore, we implement a simple probabilistic decision rule to identify regions with statistically significant regulatory behaviour. Our aim is to demonstrate that with simple assumptions about the data-generating mechanisms and decision rules, we can provide more consistent estimates of linear associations between epigenomic quantities.

Apr 28 2020 -

ANC Workshop - 28/04/2020

Christos Maniatis