The problem. Once you have cell types from a single-cell atlas, the next question is which ones signal to which — but a receptor is often a multi-subunit complex, and naïvely pairing single genes misrepresents the biology. You need a reference that knows a functional receptor requires all its subunits expressed, and a way to say an interaction is enriched rather than incidental.

The idea. CellPhoneDB is two things: a hand-curated database of ligands, receptors, and their interactions that respects heteromeric complex structure, and a statistical framework that permutes cell-type labels to test whether a ligand–receptor pair is specifically enriched between two cell types. Only when every subunit is expressed does the complex count, so the calls reflect assembled receptors, not lone transcripts.

Why it matters. This is the on-ramp for the whole cell–cell communication field, and it reframes an atlas from a catalogue of cells into a network of interactions — exactly the shift the STU cares about when a tissue’s function is the conversation between its cell types. It pairs naturally with the clustering and annotation methods from earlier days: cluster, label, then ask who signals.

Verdict. Foundational and heavily used; its weakness is that expression co-occurrence isn’t spatial proximity, which is what pushes the field toward the spatially-aware methods that follow. Read it as the vocabulary of cell–cell signaling, the dictionary the later methods speak from.