Why Science? | CellPhoneDB v5, Multimeric Complexes and Cell-Communication Specificity Evidence

Section 1 of 36

1. Begin with curated molecular compatibility

CellPhoneDB asks which annotated cell populations express compatible ligands and receptors from a curated human interaction repository. It makes molecular architecture visible, especially when ligands or receptors contain several subunits. Compatibility is the beginning of investigation, not its conclusion.

Contents · Next section

Section 2 of 36

2. Respect multimeric complexes

A functional complex may require every specified component. Collapsing it to the most visible gene can create false positives. Show subunit expression separately and state the rule used to combine them, particularly when one component is sparse or near the detection threshold.

Contents · Previous section · Next section

Section 3 of 36

3. Summarise expression by cell type

Mean expression helps show whether an interaction is supported in source and target groups. Means can be driven by a minority of cells or depth differences, so pair them with expressing-cell fractions, sample distributions and relevant quality checks.

Contents · Previous section · Next section

Section 4 of 36

4. Ask about statistical specificity

Permutation-based analysis evaluates whether an interaction is more specific to a cell-type pair than expected after relabelling under the method’s procedure. A small p-value addresses specificity within that setup; it does not measure effect size, binding strength or clinical importance.

Contents · Previous section · Next section

Section 5 of 36

5. Use differential-expression queries deliberately

CellPhoneDB v5 supports user-tailored analyses, including approaches that incorporate differentially expressed genes. The contrast, model and threshold used to generate that gene list become part of the communication result and must be documented.

Contents · Previous section · Next section

Section 6 of 36

6. Add microenvironment information

Spatially derived microenvironments can restrict candidate interactions to cell types that plausibly co-occur. This is a valuable biological filter, but its resolution and segmentation matter. Co-location makes contact or diffusion possible; it still does not prove signalling.

Contents · Previous section · Next section

Section 7 of 36

7. Use CellSign as a prioritisation layer

Transcription-factor activity can help prioritise interactions whose downstream regulatory context appears compatible with receptor activity. The extra layer is evidence integration, not causal closure. Record how activities were estimated and whether they were measured in matched cells.

Contents · Previous section · Next section

Section 8 of 36

8. Build a human-gene manifest

Record gene identifier system, organism, CellPhoneDB version, interaction database release, cell annotations, specimens, conditions and every threshold. CellPhoneDB is designed around human interactions, so any cross-species mapping requires explicit justification and audit.

Contents · Previous section · Next section

Section 9 of 36

9. Check identifiers and duplicates

Resolve Ensembl symbols, aliases and duplicated genes before analysis. A failed mapping can silently remove a subunit, while an obsolete symbol can hide a valid pair. Export mapped and unmapped identifiers rather than reporting only retained features.

Contents · Previous section · Next section

Section 10 of 36

10. Audit annotations before permutations

Permutation results inherit the cell-type labels being shuffled. Review mixed clusters, doublets, rare populations and annotation uncertainty first. Repeat important interactions after merging unstable subtypes to learn whether specificity is real or label-dependent.

Contents · Previous section · Next section

Section 11 of 36

11. Separate cells from specimens

Ten thousand cells from one person are not ten thousand biological replicates. Keep specimen identity in every table and compare pair support across individuals. Pooled permutations can describe the pooled atlas but cannot establish population-level reproducibility.

Contents · Previous section · Next section

Section 12 of 36

12. Inspect every complex component

For multimeric ligands and receptors, display subunit detection, abundance and specimen prevalence. A weakest-link rule protects against incomplete complexes, but transcript dropout can also hide real biology. Mark this tension rather than choosing whichever rule gives the nicer network.

Contents · Previous section · Next section

Section 13 of 36

13. Define microenvironments independently

If microenvironments come from spatial data, explain coordinate system, neighbourhood scale and whether the same genes informed both cell labels and interaction claims. Avoid defining a niche specifically to retain a favoured interaction.

Contents · Previous section · Next section

Section 14 of 36

14. Archive database and analysis assets

Preserve input counts, metadata, annotation hierarchy, identifier map, database files, parameters, random seed, permutation count, expression means, p-values, differential results, microenvironments, TF activities and visualisation exports.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented complex audit table

This fictional table is a reasoning exercise, not CellPhoneDB output.

Contents · Previous section · Next section

Section 16 of 36

16. Read means and p-values together

Mean expression and permutation p-values answer different questions. A high mean may be non-specific; a specific pair may have modest expression. Present both, plus detection fractions and specimen prevalence, before deciding which candidates deserve experiments.

Contents · Previous section · Next section

Section 17 of 36

17. Avoid a p-value leaderboard

Ranking solely by p-value rewards sample size and label structure while hiding magnitude and reproducibility. Define a prioritisation rule that includes complex completeness, expression, specificity, biological context and independent evidence.

Contents · Previous section · Next section

Section 18 of 36

18. Interpret differential interactions narrowly

A differential-expression-informed result indicates compatibility with a declared contrast. It does not automatically mean communication increased between conditions. Compare raw expression, cell abundance, specimen effects and matched interaction estimates.

Contents · Previous section · Next section

Section 19 of 36

19. Treat CellSign as corroboration

A receptor-linked TF activity can make an interaction more coherent, especially when derived from another modality. Yet many pathways converge on the same TFs. Use CellSign to prioritise experiments, not to state that one ligand uniquely caused the activity.

Contents · Previous section · Next section

Section 20 of 36

20. Use microenvironments as constraints

Restricting candidate pairs to plausible niches reduces impossible long-range combinations. Show how results change with and without the constraint and across reasonable neighbourhood scales. If the conclusion exists only in one hand-picked niche, say so.

Contents · Previous section · Next section

Section 21 of 36

21. Model abundance as a confounder

Cell-type proportions often change with condition and sampling. Rare groups yield unstable means and permutations; abundant groups generate many opportunities. Report counts and use specimen-level analysis so composition is not mistaken for altered signalling.

Contents · Previous section · Next section

Section 22 of 36

22. Audit database versions

CellPhoneDB v5 expanded the resource, including interactions involving non-peptidic ligands described in its protocol. Freeze the actual release and compare headline findings with a high-confidence or earlier subset where informative.

Contents · Previous section · Next section

Section 23 of 36

23. Audit expression thresholds

A detection threshold decides whether a group participates. Repeat conclusions over a small prespecified range and report interactions that cross the boundary. Threshold-sensitive candidates are useful exploratory leads but weak headline evidence.

Contents · Previous section · Next section

Section 24 of 36

24. Audit annotation resolution

Fine subtypes can make an interaction appear specific because the same cells were split into small categories. Re-run at a coarser biological level and compare. Specificity that survives both resolutions is easier to trust.

Contents · Previous section · Next section

Section 25 of 36

25. Audit microenvironment scale

A niche defined at 20 micrometres means something different from a broad anatomical region. Test plausible scales in physical units and show which cell-type pairs become possible. Do not describe regional co-residence as direct contact.

Contents · Previous section · Next section

Section 26 of 36

26. Validate beyond RNA

Use protein imaging, receptor localisation, ligand measurements, proximity assays or perturbation as appropriate. For complexes, verify required subunits. RNA-based compatibility becomes much stronger when the molecular assembly and downstream response agree.

Contents · Previous section · Next section

Section 27 of 36

27. Compare another method fairly

Run a second communication method and a simple expression baseline on the same filtered cells and resource where possible. Examine pair-level concordance, not just pathway names. Differences often expose threshold, complex or specificity assumptions.

Contents · Previous section · Next section

Section 28 of 36

28. Design a decisive follow-up

Select one complete, specific, niche-supported complex and one attractive but incomplete complex. Measure or perturb the missing subunit and downstream response. A design that can falsify the ranking is more valuable than another circular visualisation.

Contents · Previous section · Next section

Section 29 of 36

29. Build a classroom model before code

Give students receptor puzzles whose pieces must all be present. They then compare a high-expression pair with a statistically specific pair and explain why the two criteria are not interchangeable. Label observations, transformations, assumptions and conclusions separately. The goal is not to imitate specialist software; it is to make every evidence hand-off visible enough for a classmate to question.

Contents · Previous section · Next section

Section 30 of 36

30. Begin with Primary Science habits

Primary Science already supplies the foundation for CellPhoneDB v5: careful observation, fair comparison, consistent records and conclusions that fit the evidence. A simple plant, light or water investigation can show why compatible parts do not prove that an interaction actually occurred.

Contents · Previous section · Next section

Section 31 of 36

31. Use PSLE Science to practise claim limits

PSLE Science asks students to connect evidence to process without leaping beyond an experiment. With CellPhoneDB v5, ask what was measured, what came from a database, what the software estimated and which new observation could separate 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 cells, organisation, molecular transport, experimental design and evaluation. CellPhoneDB v5 turns those ideas into a contemporary data problem. The useful lesson is disciplined interpretation, not memorising a software menu.

Contents · Previous section · Next section

Section 33 of 36

33. Let Computing support the biology

Computing contributes tables, graphs, algorithms, ranking, version control and reproducibility. Biology supplies specimens, mechanisms and validation. CellPhoneDB v5 shows why correct code is necessary but insufficient: a flawless program can still analyse confounded samples or unsuitable labels.

Contents · Previous section · Next section

Section 34 of 36

34. Choose opportunities by verified fit

For school choices and science enrichment, verify current programmes on official school and MOE pages. Prioritise programmes where students can handle data and experimental controls, and verify every current offering from official sources. A fashionable mention of genomics or AI does not guarantee sustained teaching, admission, mentorship or a particular career outcome.

Contents · Previous section · Next section

Section 35 of 36

35. See connected career families

The reasoning in CellPhoneDB v5 appears in molecular biology, immunology, genomics, pathology, data curation, computational biology and translational research. Routes may pass through polytechnic, junior college, university or continuing education, with different blends of biology, statistics, computing and communication. Requirements change, so check the responsible institution directly.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a bounded scientific claim

A defensible conclusion states exactly what CellPhoneDB v5 prioritised under which data, database, settings and specimens. CellPhoneDB v5 prioritises curated, expression-supported cell-pair interactions under declared statistical and contextual rules; it does not directly observe binding, complex assembly or causal response. Report uncertainty, sensitivity, replication, negative results and independent validation together; those boundaries make a claim more useful.

CellPhoneDB’s special contribution is molecular architecture. Build every headline result from the complex upward: list all ligand and receptor subunits, their expression fractions, their specimen prevalence and the rule that converted them into a complex score. A family label or a single highly expressed chain should never stand in for an assembled receptor.

Keep the resource and the algorithm separate in the report. The curated repository decides which interactions can be tested; expression summaries and statistical procedures decide which eligible interactions are prioritised. Save both versions. A new database release can add candidates even when the biological data never change.

Human gene identifiers need exact handling. Retain a table of submitted identifiers, accepted symbols, aliases, duplicates and unmapped entries. Check whether the missing genes belong disproportionately to complexes or newer ligand classes. Silent mapping loss can produce a false biological absence.

Permutation p-values do not mean “chance that the interaction is false.” They quantify how unusual the observed cell-pair specificity is under the method’s relabelling procedure. Present the number of permutations, group sizes, expression mean and effect context. Very small groups can create unstable specificity, while huge groups can create tiny p-values for modest patterns.

The v5 protocol describes a larger repository, including interactions involving non-peptidic ligands, and new ways to tailor analyses. This breadth is exciting, but different mediator classes have different biology. Peptides, steroid hormones, neurotransmitters and other molecules should not all inherit the same spatial range, synthesis assumptions or validation assay.

Microenvironment filters can rescue plausibility by excluding cell types that do not share tissue context. Define these niches without reference to the desired interaction, state resolution and distance, and show sensitivity to plausible boundaries. A broad compartment and a contact-scale neighbourhood answer different biological questions.

CellSign brings downstream transcription-factor activity into prioritisation. Show whether the TF information comes from matched chromatin data, inferred activity or prior knowledge. Many receptors converge on the same regulators, so agreement supports coherence without uniquely identifying one ligand–receptor route.

Did you know? A heteromeric receptor can fail functionally when one required subunit is absent even if every other chain is abundant. This is why CellPhoneDB’s complex-aware design matters—and why transcript dropout creates a genuine interpretive tension rather than a simple yes-or-no rule.

Use differential-expression-informed analyses with care. State the model, covariates, minimum effect and multiple-testing rule that produced the input list. An interaction containing a differentially expressed component is not automatically a differentially active communication channel. Composition, state and detection can all change the list.

Sampling balance affects both means and permutations. Display the number of cells and specimens for every group, downsample as a sensitivity analysis when imbalance is extreme, and keep specimen identity visible. Never report pooled significance as if it represented independent people or animals.

Control ambient RNA and doublets. Secreted ligands can appear broadly because free RNA contaminated droplets, while doublets can create impossible co-expression of two cell types. Use suitable quality-control diagnostics and demonstrate that headline sources are not owned by these artefacts.

A useful baseline is a complete-complex expression rule without permutation. Another is the same algorithm with a high-confidence resource subset. These comparisons reveal whether specificity statistics and database breadth add stable information. Agreement is reassuring; disagreement tells the reader which assumption matters.

Validation should follow the molecular route. Verify that every required protein is present in the relevant cellular compartment, establish that source and receiver share an appropriate niche, perturb the ligand or receptor and measure an expected downstream response. Each step closes a distinct gap between RNA compatibility and communication.

Publish negative interactions intentionally. Include a highly expressed but non-specific pair, a specific pair with one missing complex subunit and a pair excluded by microenvironment. These examples teach how the criteria work and protect the report from presenting only successes.

Archive the database release, user additions, analysis mode, threshold, permutation seed and count, microenvironment file, TF-activity source, result tables and CellPhoneDBViz assets. An interactive figure is a doorway, not the scientific record. The underlying rows must remain downloadable and interpretable.

A bounded conclusion might say: “CellPhoneDB v5 prioritised a complete multimeric interaction between annotated source and receiver groups, with cell-pair specificity, specimen-level expression and independent niche support.” It should not say the complex assembled, bound or caused the receiver programme unless those events were measured.

Design a follow-up that can overturn both the complex and the context. Test a required subunit and move the experiment across a niche boundary or co-culture arrangement. If signalling persists without the subunit or outside the proposed context, the computational explanation needs revision.

Before publication, run a language audit as well as a numerical one. Replace “cells communicate” with “the analysis prioritises a candidate interaction” wherever the evidence is transcript-only. Reserve stronger verbs for measurements of protein, binding, activation or response. Precise language is not timid; it tells a future experiment exactly which inferential step remains open.

Contents · Previous section · Continue to the Science Learning Hub