Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data

TLDR

Single‑cell transcriptomics has spurred interest in inferring cell‑cell communication, leading to many computational tools that combine interaction resources with predictive methods, yet the effect of choosing different resources and methods remains largely unknown. The study systematically compares 16 cell‑cell communication inference resources and 7 methods, plus the consensus between their predictions. The authors evaluate all combinations of methods and resources and provide the LIANA framework as an open‑source interface to these tools. The comparison revealed that resources differ in unique interactions, overlap, and pathway coverage; method–resource combinations strongly shape predicted interactions, and predictions generally agree with spatial colocalisation, cytokine activity, and receptor protein abundance data.

Abstract

Abstract The growing availability of single-cell data, especially transcriptomics, has sparked an increased interest in the inference of cell-cell communication. Many computational tools were developed for this purpose. Each of them consists of a resource of intercellular interactions prior knowledge and a method to predict potential cell-cell communication events. Yet the impact of the choice of resource and method on the resulting predictions is largely unknown. To shed light on this, we systematically compare 16 cell-cell communication inference resources and 7 methods, plus the consensus between the methods’ predictions. Among the resources, we find few unique interactions, a varying degree of overlap, and an uneven coverage of specific pathways and tissue-enriched proteins. We then examine all possible combinations of methods and resources and show that both strongly influence the predicted intercellular interactions. Finally, we assess the agreement of cell-cell communication methods with spatial colocalisation, cytokine activities, and receptor protein abundance and find that predictions are generally coherent with those data modalities. To facilitate the use of the methods and resources described in this work, we provide LIANA, a LIgand-receptor ANalysis frAmework as an open-source interface to all the resources and methods.