We aimed to develop a flexible and robust integration strategy to integrate data from single-cell sequencing experiments where different modalities are measured (‘single-modality datasets’). The fundamental challenge is that different single-modality datasets measure different sets of features. For example, scRNA-seq measures the expression level of individual transcripts, while scATAC-seq or scBS-seq measure DNA accessibility or methylation levels (Fig. 1a). Previously proposed methods from our group and others19-21 attempt to convert one set of features into another, for example, taking the gene-body sum of ATAC-seq signal (or the inverse of the DNA methylation levels), as a proxy for transcriptional output. While this conversion facilitates downstream integration, it assumes a strict and simplistic biological relationship between modalities that may not hold true, particularly in developing or transitioning systems.
