connectomeDB2025: a rigorously curated, multi-species resource of experimentally supported ligand–receptor interactions

Abstract Inferring cell–cell communication networks is now a cornerstone of single-cell RNA-seq and spatial transcriptomics data analysis, relying critically on reference catalogues of experimentally supported ligand–receptor interactions. Here, we present the updated, rigorously curated connectomeDB, an open-access database of peptide-based ligand–receptor pairs comprising 3579 vertebrate interactions supported by primary experimental evidence from 2803 research articles. By critically reviewing all putative ligand–receptor pairs from connectomeDB2020, CellChatDB v2, CellPhoneDB v5, CellTalkDB, ICELLNET v2, and LIANA+, we first removed over 2900 misclassified or unsupported interactions lacking primary-literature evidence. We then expanded the resulting verified dataset through AI-assisted literature mining and manual curation, adding 827 pairs and 718 supporting articles absent from other databases, including 264 pairs first described since 2020. connectomeDB2025 contains 5429 evidence links (“triplets”), each connecting a ligand–receptor pair to a specific publication, collectively providing at least one source of primary experimental evidence for each interaction. Notably, 2359 of these triplets are exclusive to connectomeDB2025, making it the most robustly supported ligand–receptor database with primary experimental evidence. The online resource (https://connectomedb.org) provides searchable, downloadable ligand–receptor lists and detailed pair summaries, enabling accurate cell–cell communication analysis across human, mouse, and 12 other vertebrate species.