Methods and interpretation boundaries

CellModelDB keeps evidence types, source lineage and interpretation limits visible rather than collapsing them into a single model claim.

Model inclusion and curation

The current release focuses on human non-tumour cell models with sufficiently defined identity and biological context for structured curation. Each record is reviewed across identity, provenance, intended target-cell or organ context, literature evidence and public-omics availability. Missing or unresolved evidence is retained explicitly. Evidence availability does not imply model validation, and the first release is designed for iterative expansion rather than exhaustive coverage.

How evidence is organized

Identity→Provenance→Literature evidence→Physiological context→Public omics→Disease-reference evidence

These remain separate evidence layers and are not automatically combined into a score.

What CellModelDB does

Identity curation

Model identity, provenance, target cell, immortalization or engineering status and explicit missingness states are retained as separate fields.

Literature evidence

Curated literature records preserve source status and stated model strengths or limitations. Missing evidence is not inferred.

Physiological context

HPA-based analyses use common Ensembl genes, Spearman similarity and target-cell elevated-gene enrichment. This is not disease fidelity.

Disease-fidelity framework

Within-dataset effects and independently defined references are required. No raw matrices are merged across studies. Current formal result count is 0.

Reference architecture

Human and mouse reference evidence are retained as separate layers. Cross-species comparisons require explicit ortholog mapping and are not inferred.

Provenance

Public assets retain release-level source identifiers and provenance metadata without exposing internal filesystem paths in the user interface.