ULMnet: inferring physical cell-cell communication networks from scRNAseq data using univariate linear models.

Area of Science:

• Single-cell genomics
• Computational biology
• Systems biology

Background:

• Single-cell RNA sequencing (scRNAseq) is crucial for understanding cellular heterogeneity.
• Tissue dissociation for scRNAseq often disrupts physical cell-cell contacts, limiting interaction inference.
• Multiplets, or cells sequenced together, may represent physically interacting cells.

Purpose of the Study:

• To develop a computational method for identifying multiplets in scRNAseq data.
• To predict the cellular composition of multiplets.
• To infer physical cell-cell interaction networks from scRNAseq data.

Main Methods:

• Developed ULMnet, a computational method using univariate linear models.
• Applied ULMnet to various scRNAseq datasets, including cell pairs, partially dissociated tissues, and healthy/cancer tissues.
• Validated ULMnet performance against FACS-sorted doublets and existing methods.

Main Results:

• ULMnet achieved high precision (~99%) and good sensitivity (~56%) in doublet prediction.
• Analysis of partially dissociated tissues revealed physical interaction networks recapitulating microanatomy.
• Cancer scRNAseq data analysis identified biologically plausible interactions validated by spatial transcriptomics.

Conclusions:

• ULMnet effectively identifies multiplets and infers physical cell-cell interactions from scRNAseq data.
• The method accurately captures biologically meaningful interactions reflecting tissue architecture.
• ULMnet offers a valuable approach to harness scRNAseq for studying physical cell contacts.