The problem. After the wave of cell–cell communication tools (CellPhoneDB, CellChat, NicheNet, and more, which I read last week), a practical worry surfaced: do they agree? Each pairs a different inference method with a different ligand–receptor resource, and if those choices drive the predictions, then any single tool’s output is hard to trust on its own.

The idea. LIANA systematically compares many CCC methods and resources on common footing, decoupling the scoring method from the interaction database so their separate effects are visible. The finding is cautionary: resources share surprisingly few interactions and cover pathways unevenly, and both the method and the resource strongly shape which interactions are called. To make comparison and consensus practical, LIANA provides a single framework that runs the methods together and aggregates them, checked against spatial colocalisation and other modalities.

Why it matters. This is the critical-evaluation counterweight to the CCC methods I already reviewed — the reminder that tool choice is a hidden variable, and that consensus beats any one caller. It’s the same benchmarking instinct as the spatial-deconvolution and integration comparisons on my list. For the STU, where niche signaling is a core readout, knowing how fragile a single method’s calls can be is essential hygiene.

Verdict. A valuable, widely-cited benchmark and a practical consensus tool; aggregation reduces but doesn’t erase resource bias. Read it as the field auditing itself — and a better default than trusting one CCC method alone.