Cell IO: a database linking single-cell transcriptomes to literature-derived knowledge of cell functions and perturbations

Single-cell transcriptome databases tell us which genes a cell type expresses, but what a cell type does, and what happens to it when a gene is perturbed, remains scattered across the literature. Cell IO (https://cell-io.riken.jp/) links these two kinds of knowledge through Cell Ontology identifiers. Its transcriptome layer re-processes 10x Chromium single- cell and single-nucleus RNA-seq experiments of human, mouse, and rat from the Sequence Read Archive with a uniform pipeline, annotates clusters with Cell Ontology terms restricted by the tissue of origin, and stores the cell type composition and UMAP embedding of every experiment (currently 14,931 experiments and 82.9 million cells). Its literature layer holds 895,258 claims extracted with a large language model and verified in a second pass from open-access abstracts, review articles, and gene perturbation studies in immunology and neuroscience, each with the verbatim sentence, highlighted entity spans, evidence strength, and ontology-normalized cell, gene, and context. The web interface lets users search cell types by name, synonym, or identifier, find the experiments in which a cell type is most abundant, inspect the composition and embedding of each experiment, read and filter the claims about a cell type by knowledge base, direction, evidence strength, and organism, and identify cell types whose functional characterization is missing from part of the literature. Cell IO is freely available without registration.