The last decade has witnessed the irresistible rise of single-cell RNA-seq, a technology which can measure simultaneously the mRNA expression levels of thousands of genes in thousands of individual cells. This has enabled an unconfounded measurement of cell states, and the discovery of new cell populations. More recently, scRNA-seq has been paired with other techniques (such as BS-seq or ATAC-seq) to measure simultaneously multiple molecular features in the same cell. The promise of these technologies is the ability to understand gene regulation at the single cell level, but, so far, the high levels of noise have prevented significant inroads in that direction. In this talk I argue that some of these effects might be remedied by taking a Bayesian statistical perspective to the modelling of noise in the data. I show that a simple hierarchical model can lead to improved estimates of correlations between genomic features, a taxing task with standard tools (Maniatis et al, PLoS CompBio 2022). I then move on to present a more structured Bayesian model which aims to infer biological pathways (topics) from joint scRNA-seq scATAC-seq protocols data.