Why Science? | CellTalkDB, Literature-Supported Ligand–Receptor Curation and Species Evidence

Section 1 of 36

1. Start with a curated catalogue

CellTalkDB supplies literature-supported ligand–receptor pairs for human and mouse. It is an evidence resource used before sample scoring begins. A listed pair says that an interaction was supported in an examined context; it does not establish expression, location or activity in the specimen now being analysed.

Contents · Next section

Section 2 of 36

2. Use the published snapshot accurately

The primary paper reported 3,398 human pairs and 2,033 mouse pairs, with 815 human ligands, 780 human receptors, 651 mouse ligands and 588 mouse receptors. Treat these as figures for the published release, not as permanent live totals or measures of biological importance.

Contents · Previous section · Next section

Section 3 of 36

3. Understand the curation pipeline

The study began with STRING protein–protein interactions, mapped proteins to gene annotations, classified potential ligands and receptors, searched PubMed and manually assessed evidence. Each transition narrows a very large candidate space and should remain visible when a record is cited.

Contents · Previous section · Next section

Section 4 of 36

4. Keep human and mouse evidence separate

The catalogue was built independently for human and mouse rather than converting every human interaction into a mouse pair. The paper reported common and species-specific records. Species assignment is therefore part of the evidence, not a decorative field that can be discarded during analysis.

Contents · Previous section · Next section

Section 5 of 36

5. Know how candidates were defined

Potential ligands were associated with secretion or extracellular distribution, while potential receptors were associated with the cell membrane or surface. These operational definitions make screening possible, but unusual signalling mechanisms and annotation errors can sit outside their boundary.

Contents · Previous section · Next section

Section 6 of 36

6. Read literature support as provenance

Potential pairs were searched with gene symbols, synonyms and protein names, then checked in full articles and supplementary materials. The existence of a supporting publication improves traceability. It does not guarantee that every experiment used the same species, tissue, molecular form or cellular state.

Contents · Previous section · Next section

Section 7 of 36

7. Separate catalogue evidence from specimen evidence

A CellTalkDB row and a single-cell expression matrix are different layers. Database evidence supports eligibility; sample expression supports presence of transcripts. Protein abundance, complex assembly, spatial opportunity, ligand release and receiver response require additional measurements.

Contents · Previous section · Next section

Section 8 of 36

8. Create a source-to-sample manifest

Record database release or access date, species, tissue, condition, donor or animal, assay, cell annotations, preprocessing and scoring tool. Preserve the original CellTalkDB identifiers. A later database update should not silently change the scientific question or the eligible pair universe.

Contents · Previous section · Next section

Section 9 of 36

9. Audit identifiers and synonyms

Gene symbols change, aliases collide and protein families contain similar names. Retain the mapping from database record to analysis identifier and flag unmapped, duplicated or ambiguous entries. Manually inspect every headline pair instead of assuming automated symbol conversion is biologically exact.

Contents · Previous section · Next section

Section 10 of 36

10. Verify the species before scoring

Match the catalogue species to the organism actually measured. If orthology is introduced later, mark it as an analyst-added inference and document the mapping source. A direct mouse evidence record and a human-derived orthologue should not receive the same provenance label.

Contents · Previous section · Next section

Section 11 of 36

11. Inspect ligand and receptor annotations

Review extracellular, membrane and functional annotations for leading records against current primary evidence. Some proteins have several cellular locations or isoforms. The ligand–receptor label should describe the molecular form relevant to the proposed route, not merely the most convenient gene name.

Contents · Previous section · Next section

Section 12 of 36

12. Freeze the eligible pair universe

Save the exact pair table used, including species and evidence identifiers. Report how many records entered before expression filtering and how many remained afterward. Without this denominator, a network can look selective even when it simply reflects a small or heavily filtered database.

Contents · Previous section · Next section

Section 13 of 36

13. Preserve cell labels and specimen keys

Communication scoring usually aggregates expression by cell groups. Check annotation certainty, doublets, rare groups and cells per specimen before scoring. Many cells from one donor improve resolution but do not create independent biological replication for a population-level claim.

Contents · Previous section · Next section

Section 14 of 36

14. Archive the evidence trail

Keep the database snapshot, mapping table, cell metadata, preprocessing, expression summaries, scoring settings, thresholds, edge tables and plotting code. For every reported interaction, retain the supporting publication and the steps that moved it from catalogue record to ranked sample hypothesis.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented provenance table

This fictional table is for evidence-reading practice, not CellTalkDB output. Compare a directly supported same-species pair, a species-specific record, an alias-sensitive mapping and a literature-supported pair absent from the specimen. Ask which next test addresses the weakest evidence link.

Contents · Previous section · Next section

Section 16 of 36

16. Read counts as coverage, not truth

A resource with more pairs explores a broader hypothesis space and can return a denser network. It is not automatically more accurate. Compare counts alongside evidence rules, species, complex handling, curation date and the biological question each resource was designed to answer.

Contents · Previous section · Next section

Section 17 of 36

17. Use shared and species-specific records

The published analysis reported common human–mouse pairs as well as records specific to each species. Shared membership can support conservation hypotheses. Species-specific membership may reflect biology, evidence density or curation history, so it needs source review before evolutionary interpretation.

Contents · Previous section · Next section

Section 18 of 36

18. Compare resources on harmonised identifiers

When comparing CellTalkDB with CellPhoneDB, SingleCellSignalR or another resource, harmonise gene symbols, direction, species and complex representation first. Record overlap and unique entries. A unique pair is a prompt to inspect provenance, not immediate proof that one catalogue is better.

Contents · Previous section · Next section

Section 19 of 36

19. Use the interface as an evidence browser

The published interface allowed browsing, searching and downloading pairs with species, orthology and supporting evidence. Use it to trace a claim back to its record. Archive the accessed data because a live interface can change after the analysis is completed.

Contents · Previous section · Next section

Section 20 of 36

20. Measure knowledge-density bias

Well-studied immune, cancer and developmental systems may contribute more verified interactions than neglected tissues. Compare cell connectivity with literature density or resource frequency. A high-degree cell may reflect annotation richness as well as broad biological communication.

Contents · Previous section · Next section

Section 21 of 36

21. Check molecular complexes separately

The original resource is centred on protein–protein ligand–receptor pairs. If a downstream scoring method represents multimeric complexes, document how individual CellTalkDB records are combined and whether all essential subunits are detected. One abundant chain should not stand for an assembled complex.

Contents · Previous section · Next section

Section 22 of 36

22. Audit the PubMed search contract

The paper searched ligand and receptor names with interaction or binding terms, including synonyms. Reproduce the logic for important records and test alternative names. Search sensitivity and naming choices determine which articles can enter manual review and which evidence remains invisible.

Contents · Previous section · Next section

Section 23 of 36

23. Audit the manual-review boundary

Three reviewers examined full text and supplementary material for candidate pairs. That is a valuable quality-control layer, yet every manual system has scope and judgement. Record the paper, experimental system, species and evidence type for headline pairs rather than citing the database alone.

Contents · Previous section · Next section

Section 24 of 36

24. Treat synonyms as an error surface

A synonym may refer to another protein, a historical name or a broad family. Verify symbols against the actual sequence, construct or reagent described in the supporting study. Alias mistakes can manufacture an interaction or assign it to the wrong species.

Contents · Previous section · Next section

Section 25 of 36

25. Publish informative absences

Distinguish not in database, mapping failure, transcript absent, receptor absent, incomplete complex, spatially incompatible, unreplicated and experimentally rejected. These states have different meanings. Collapsing them into one ‘no communication’ category discards useful evidence and hides where the pipeline failed.

Contents · Previous section · Next section

Section 26 of 36

26. Add spatial and protein evidence

For leading pairs, confirm that source and receiver cells can encounter each other and that the relevant proteins or molecular complexes are present. Contact-dependent, diffusible and matrix-mediated routes require different spatial expectations. Expression compatibility alone cannot decide the correct geometry.

Contents · Previous section · Next section

Section 27 of 36

27. Design a discriminating perturbation

Choose two candidate routes that predict different early receiver events. Block the ligand, receptor or required component under matched timing, then measure proximal response before late phenotype. Include a result that would reject both mechanisms rather than validating only the favourite edge.

Contents · Previous section · Next section

Section 28 of 36

28. Write a bounded catalogue claim

Prefer: ‘A CellTalkDB record supplied literature-supported eligibility for a pair that was expressed and replicated in this dataset.’ Avoid saying the cells communicated until transfer, receptor activation and response have been tested. The boundary makes the database more scientifically useful.

Contents · Previous section · Next section

Section 29 of 36

29. Build a classroom curation exercise

Give students four short evidence cards and ask whether each supports ligand identity, receptor identity, binding, species or tissue relevance. Then build one database row together. The exercise shows that a catalogue is an argument assembled from sources, not a list delivered without assumptions.

Contents · Previous section · Next section

Section 30 of 36

30. Begin with Primary Science habits

Primary Science builds observation, classification, fair comparison and careful recording. CellTalkDB extends the same habits: define categories, keep evidence attached to claims and separate what is observed from what is inferred. The software is advanced, but the reasoning discipline begins early.

Contents · Previous section · Next section

Section 31 of 36

31. Use PSLE Science to practise claim limits

PSLE Science asks students to link evidence to a process without jumping beyond the experiment. With a ligand–receptor record, ask what was directly tested, what belongs to the database, what appears in the new sample and what experiment would distinguish two explanations.

Contents · Previous section · Next section

Section 32 of 36

32. Connect Secondary and O-Level Science

Secondary Science and O-Level Biology develop ideas about cells, proteins, coordination, experimental design and evaluation. A curated interaction database turns those ideas into a modern evidence problem. The valuable skill is reading the chain from molecular claim to dataset, not memorising record counts.

Contents · Previous section · Next section

Section 33 of 36

33. Let Computing support reproducibility

Computing contributes identifiers, tables, search rules, version control and reproducible pipelines. Biology supplies molecular definitions and validation. CellTalkDB shows why both are required: a technically correct join can still connect the wrong alias, species or molecular form.

Contents · Previous section · Next section

Section 34 of 36

34. Choose opportunities by verified fit

For science enrichment and school choices, check current programme details on official school and MOE pages. Look for sustained opportunities in biology, data handling, source evaluation and research practice. A brochure mention of genomics or AI does not guarantee admission, mentorship or a particular outcome.

Contents · Previous section · Next section

Section 35 of 36

35. See connected career families

The work connects to bioinformatics, database curation, molecular biology, scientific publishing, knowledge engineering and research software. Routes may pass through polytechnic, junior college, university or continuing education. Requirements change, so verify each institution and role directly rather than promising one fixed pathway.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a bounded scientific claim

A defensible conclusion states which CellTalkDB release, species, identifiers and evidence records defined the eligible pairs, then reports specimen expression, replication and independent validation separately. CellTalkDB strengthens provenance; it does not convert a literature-supported catalogue entry into observed signalling in a new tissue.

CellTalkDB is most useful when every database row is treated as a documented proposition rather than a verdict. Keep the ligand, receptor, species, source record, supporting publication and database version together. Then a reader can distinguish what the literature established from what the new specimen still needs to show.

Did you know? The published CellTalkDB paper reported 3,398 human ligand–receptor pairs and 2,033 mouse pairs. Those counts are properties of the authors’ published curation snapshot and rules. They are not counts of active routes in a tissue, and they should not be silently updated or transferred between species.

The curation pipeline began from protein–protein interactions and then narrowed candidates using ligand and receptor annotations. This is a strong reminder that a database is produced by successive filters. Record the starting universe, the eligibility rules and every later manual decision. Otherwise, a missing interaction can be mistaken for negative biological evidence when it may simply fall outside the catalogue’s construction.

Species must remain an explicit field. A human-supported pair can motivate a mouse experiment, but orthology does not guarantee identical expression, binding, tissue context or regulatory use. Map identifiers transparently, keep one-to-many orthologues visible and label transferred hypotheses separately from interactions curated directly in that species.

Synonyms are another quiet source of error. Gene symbols change, proteins may be referred to by family names and older papers can use historical aliases. Preserve the database identifiers and the exact symbols used in the specimen. An auditable mapping table prevents a plausible-looking pair from joining the wrong molecular entities.

Manual review improves traceability, yet published evidence is uneven. Some pathways, diseases and tissues attract far more experiments than others. Compare candidate prominence with literature density. A well-studied ligand may appear central partly because it has been easier to annotate, while a less studied route may be absent without being biologically impossible.

Database membership supplies prior plausibility, not specimen evidence. At minimum, check ligand and receptor expression in the appropriate source and receiver populations, prevalence across biological specimens and any required receptor components. Next examine anatomical opportunity. Only after those gates should the route advance to protein, proximal signalling and perturbation tests.

Expression thresholds need sensitivity checks. A candidate just below a cut-off is not fundamentally different from one just above it. Publish detection fractions and continuous expression summaries, then repeat key rankings across a planned threshold range. Stable conclusions should not depend on one convenient boundary.

Multimeric biology deserves special handling even when the database row appears simple. A receptor may require several chains, cofactors or processing steps. List required components and show which were observed. One abundant subunit should never stand in for a functional complex unless the model and evidence explicitly justify that simplification.

Source and receiver labels can also be uncertain. Doublets, ambient RNA and transitional states may create apparent co-expression or self-signalling. Repeat the analysis with conservative quality filters and at a broader cell hierarchy. A route that survives those changes is more credible than one owned by a fragile subcluster.

Use spatial evidence according to mechanism. Secreted factors, extracellular-matrix signals and contact-dependent interactions operate across different distances. A single neighbourhood radius is not a universal validation rule. State the mediator class, tissue compartment and distance assumption before testing colocalisation.

The most useful comparison is not simply how many pairs another resource contains. Hold the specimen, cell labels and scoring rule fixed, then compare the shared and resource-specific candidates. Shared high-confidence routes form a robust core; resource-specific routes become transparent follow-up questions about curation scope and evidence policy.

Negative controls should probe different weaknesses. Test incompatible anatomical compartments, shuffled receptor identities with preserved expression frequency, species-transferred pairs and curated pairs whose proteins are experimentally absent. Each control addresses a separate source of false confidence.

Validation should progress from cheap to discriminating. Confirm labels and RNA first, then receptor or ligand protein, then a proximal receiver event, and finally timed perturbation and rescue where practical. If the early gate fails, do not spend an elaborate experiment trying to rescue the database prediction.

Archive the exact catalogue release or export with the analysis. If a future version adds or removes a row, describe that as a knowledge-base change. The tissue did not change retrospectively. This simple version rule makes longitudinal and cross-study comparisons scientifically honest.

A strong report includes at least one supported route, one catalogue-only route, one species-transfer hypothesis and one negative control. Showing all four teaches readers how the evidence ladder works and prevents the successful example from making every database row look equally established.

For students, this is a beautiful lesson in classification, sources and fair testing. Primary Science asks learners to observe carefully; PSLE Science develops evidence-based explanation; Secondary and O-Level Science deepen variables, models and biological systems. CellTalkDB extends the same habits into a real research catalogue.

The bounded conclusion is practical: CellTalkDB can supply a literature-supported, species-labelled shortlist for communication analysis. It cannot by itself establish expression, molecular contact, direction, activity or causation in a new specimen. Its value grows when provenance and missing evidence remain visible.

Contents · Previous section · Continue to the Science Learning Hub