Probabilistic modelling of single-cell bisulfite sequencing data with MethylVI

Abstract

Single-cell bisulfite sequencing (scBS-seq) has enabled the study of methylomic heterogeneity in complex tissues at an unprecedented resolution. Yet, fully leveraging these rich datasets poses a substantial challenge, as genuine epigenetic variability is confounded by a unique set of technical factors (for example, sparse coverage of cytosines). Here we present methylation variational inference (MethylVI), a probabilistic modelling framework for scBS-seq that explicitly disentangles meaningful biological signals from confounding nuisance variations. A key strength of our model is its flexibility: MethylVI supports multiple core analysis tasks out-of-the-box, including dimensionality reduction, integration of data from different scBS-seq protocols and identification of differentially methylated genes. Moreover, our implementation readily integrates with other single-cell analysis tools, which we demonstrate by extending MethylVI using previously proposed frameworks for single-cell reference atlas mapping and multiomic modelling. MethylVI enhances analyses of single-cell bisulfite sequencing methylomic data via a deep generative model that accounts for the unique technical and biological sources of variability in this data modality.