Cells in tissues communicate in many ways, by sending signals over distance or by directly touching neighboring cells. Single-cell RNA sequencing has become a standard tool for studying these interactions, but it has an important limitation. To sequence individual cells, tissues must be dissociated, and this process breaks apart physical cell contacts, making it hard to identify which cells were actually interacting in the tissue.

Researchers at University College Dublin describe a computational approach that turns this limitation into an opportunity. During tissue dissociation, some physically attached cells remain stuck together and are sequenced as one unit. These events are often called multiplets or doublets and are usually treated as technical artifacts and removed. The authors instead show that these multiplets can provide valuable clues about real physical cell-cell interactions.

The new method, called ULMnet, uses univariate linear models to identify multiplets in scRNAseq datasets, predict which cell types they contain, and reconstruct physical interaction networks. By analyzing how gene expression patterns combine when two cells are sequenced together, ULMnet infers which cell types were likely in direct contact before dissociation.

The researchers tested ULMnet across several datasets, including experimentally sorted cell pairs, partially dissociated tissues, and standard RNA sequencing datasets from healthy and cancer samples. The method showed very high precision when identifying known cell pairs and performed as well as or better than existing approaches.

When applied to partially dissociated tissues, ULMnet recovered interaction networks that matched known tissue structures, such as expected spatial relationships between immune and stromal cells. In cancer datasets, the inferred interactions aligned with biologically plausible tumor microenvironment contacts and were supported by independent spatial transcriptomics data.

Overall, the work demonstrates that RNA sequencing data can be used not only to study gene expression but also to infer physical cell-cell interactions. By rethinking how multiplets are handled, ULMnet adds a new layer of biological insight to single-cell analysis.

Availability – The ULMnet R package is under development and will soon be available on Bioconductor. However it can currently be installed and accessed via https://github.com/Sidex71/ULMnet.