Why Science? | Domino, Transcription-Factor Activation and Ligand–Receptor Network Evidence

Section 1 of 36

1. Begin at the receiver

Domino reverses the usual storytelling direction. It first estimates transcription-factor activity in receiving cells, then asks which receptors track that activity and which ligands could connect to those receptors. This is a disciplined hypothesis generator, not a measurement of a message travelling from one cell to another.

Contents · Next section

Section 2 of 36

2. Keep the 2021 study boundary

The method was introduced in a 2021 Nature Biomedical Engineering paper studying signalling networks associated with implanted biomaterials. Results reported there belong to those datasets and analyses. A new tissue, disease or developmental system requires a new analysis and its own biological validation.

Contents · Previous section · Next section

Section 3 of 36

3. Estimate transcription-factor activity

The published workflow used SCENIC-derived transcription-factor activity rather than relying only on transcription-factor RNA abundance. That helps connect a receiver state to regulatory evidence. It also means every receptor association inherits assumptions from gene-regulatory-network construction, target selection and activity scoring.

Contents · Previous section · Next section

Section 4 of 36

4. Correlate receptors with TF activity

Domino searches for receptors whose expression correlates with active transcription factors. Correlation can prioritise a receptor that plausibly sits upstream of a receiver programme. It cannot prove that the receptor caused the programme; common cell state, treatment, batch or cell composition can move both variables together.

Contents · Previous section · Next section

Section 5 of 36

5. Map receptors back to ligands

Eligible ligand partners are then retrieved through a ligand–receptor knowledge resource, including the CellPhoneDB resource used in the published workflow. The mapping supplies molecular plausibility. It does not show that the ligand was secreted, reached the receptor or activated it in the analysed specimen.

Contents · Previous section · Next section

Section 6 of 36

6. Construct cluster-level networks

Candidate ligands, receptors and transcription factors can be organised by cell cluster to create an interpretable network. Cluster summaries make complex data readable, but they can hide rare cells, transitional states and donor differences. Always retain the cell-level and specimen-level evidence behind a neat diagram.

Contents · Previous section · Next section

Section 7 of 36

7. Separate reconstruction from observation

RNA counts are observed through an assay; inferred TF activity, receptor correlations and ligand mappings are modelled layers. Label each layer explicitly. The scientific strength of Domino is the connected evidence chain, not permission to describe every reconstructed edge as direct communication.

Contents · Previous section · Next section

Section 8 of 36

8. Create a versioned manifest

Record specimen, condition, assay, preprocessing, cell annotations, SCENIC settings, TF regulons, receptor universe, ligand–receptor database and Domino implementation. Preserve software versions and access dates. The original Domino code and the current dominoSignal project must not be silently treated as the same release.

Contents · Previous section · Next section

Section 9 of 36

9. Prepare counts, z-scores and labels

The current dominoSignal documentation describes inputs including count data, z-scored counts, cluster labels, TF activity and ligand–receptor pairings. Check that matrices share cell identifiers and genes. Document filtering, normalisation and scaling because each changes which patterns can be correlated.

Contents · Previous section · Next section

Section 10 of 36

10. Audit the TF activity layer

Inspect regulon size, target support and activity distributions for each headline transcription factor. Repeat key conclusions under reasonable SCENIC settings or an independent activity method. A familiar TF name is not evidence that its regulon is well identified in every cell type.

Contents · Previous section · Next section

Section 11 of 36

11. Freeze the ligand–receptor resource

Save the exact pair table, species, complex rules and identifier mapping. A database update can add or remove eligible ligands while the biological sample stays unchanged. Publishing the pair universe makes that analytical change visible and keeps the result reproducible.

Contents · Previous section · Next section

Section 12 of 36

12. Separate original Domino and dominoSignal

The Elisseeff-Lab Domino repository is archived and preserves the published lineage; dominoSignal is the actively developed successor with its own documentation and releases. State which one produced the result. New features or changed defaults in a successor cannot be attributed retrospectively to the 2021 paper.

Contents · Previous section · Next section

Section 13 of 36

13. Protect specimen independence

Cells are nested within donors, animals or experimental samples. Estimate associations inside biological specimens where possible and report prevalence across them. Thousands of cells from one implant do not create thousands of independent implants, however impressive the heat map appears.

Contents · Previous section · Next section

Section 14 of 36

14. Archive every intermediate

Retain expression matrices, metadata, regulons, TF activities, receptor correlations, pair mappings, cluster networks, thresholds, random seeds and plotting code. A final network image cannot reveal which input or decision created an edge, so it is not an adequate reproducibility record.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented evidence table

This fictional table is for reasoning practice, not Domino output. Notice how the same attractive network edge can have a different weakest link. Choose the next experiment that addresses that link rather than repeating the strongest evidence.

Contents · Previous section · Next section

Section 16 of 36

16. Read TF activity as a model output

A high score means the observed expression pattern resembles the selected regulon’s expected activity. It is not direct protein localisation, phosphorylation or DNA binding. Use orthogonal assays when a mechanistic conclusion depends on the transcription factor itself.

Contents · Previous section · Next section

Section 17 of 36

17. Read receptor correlations conditionally

Publish correlation magnitude, uncertainty, number of cells and specimen support rather than a yes-or-no edge alone. Stratify by relevant covariates. A correlation that disappears within specimens or cell states may reflect pooling rather than a robust receptor–programme relationship.

Contents · Previous section · Next section

Section 18 of 36

18. Read ligands as candidates

A database-compatible ligand completes the proposed chain but remains a candidate source signal. Report which cell group expresses it, the detection fraction and specimen prevalence. Confirm secretion, processing or membrane presentation according to the ligand’s actual biology.

Contents · Previous section · Next section

Section 19 of 36

19. Interpret network modules carefully

Groups of connected ligands, receptors and TFs can highlight coordinated programmes. Name a module from several stable members, not one famous molecule. Compare module membership across seeds, thresholds and specimens before assigning a biological label.

Contents · Previous section · Next section

Section 20 of 36

20. Add time when direction matters

A receiver TF programme observed at the same endpoint as ligand RNA does not establish order. A useful design measures ligand availability, an early receptor-proximal event and later TF or target activity at justified intervals. Temporal order narrows alternatives without proving causality by itself.

Contents · Previous section · Next section

Section 21 of 36

21. Add spatial opportunity

Test whether source and receiver cells occupy compatible compartments and distances. Contact signals, short-range paracrine factors and broadly diffusible molecules require different spatial expectations. Co-localisation supports opportunity, not transfer; separation can be a strong falsifying observation.

Contents · Previous section · Next section

Section 22 of 36

22. Audit correlation thresholds

Repeat the receptor–TF analysis across a planned range of correlation and significance cut-offs. Report continuous values for leading edges. A route that exists only at one arbitrary threshold should be described as fragile rather than promoted as a definitive network.

Contents · Previous section · Next section

Section 23 of 36

23. Audit SCENIC sensitivity

Change reasonable regulon, filtering and activity settings while keeping the biological question fixed. Track whether the same TFs and receptor links remain. Sensitivity is particularly important when a transcription factor has few detected targets or a regulon dominated by one highly variable gene.

Contents · Previous section · Next section

Section 24 of 36

24. Audit database sensitivity

Compare compatible ligand–receptor resources after harmonising species, identifiers, direction and complex handling. Shared routes form a robust core; resource-specific routes reveal differences in curation scope. Do not select the database merely because it produces the most exciting network.

Contents · Previous section · Next section

Section 25 of 36

25. Publish broken chains

Report candidates that fail because the ligand is absent, the receptor lacks correlation, TF activity is inconsistent, cells are spatially separated or specimens disagree. These informative negatives show where the model stopped and prevent a network figure from looking universally confirmed.

Contents · Previous section · Next section

Section 26 of 36

26. Measure proteins and proximal response

For a leading route, confirm ligand and receptor protein in the appropriate cells, then assay an early receptor or pathway event before a broad transcriptional endpoint. This sequence tests the weakest causal steps instead of treating downstream similarity as sufficient proof.

Contents · Previous section · Next section

Section 27 of 36

27. Design a discriminating perturbation

Block the nominated ligand or receptor under matched timing and controls, then measure the predicted early receiver event and TF programme. Include an alternative ligand or receptor prediction. A useful experiment can reject the favoured route, distinguish competitors and reveal compensation.

Contents · Previous section · Next section

Section 28 of 36

28. Write a bounded Domino claim

Prefer: ‘Domino prioritised a ligand–receptor–TF chain supported by receiver activity, receptor correlation and a curated pair resource across three specimens.’ Avoid: ‘Domino proved that the source cells activated the receiver.’ The first statement is precise and testable.

Contents · Previous section · Next section

Section 29 of 36

29. Turn the chain into a classroom investigation

Give students cards for ligand expression, receptor correlation, TF activity, spatial location and perturbation. Ask them to build the strongest allowed claim and identify the missing evidence. The activity makes inference layers visible and rewards careful scientific language.

Contents · Previous section · Next section

Section 30 of 36

30. Begin with Primary Science habits

Observation, classification, fair comparison and accurate recording are the roots of this workflow. A younger learner can practise separating what was measured from what was inferred long before learning single-cell genomics. The habit is already genuine science.

Contents · Previous section · Next section

Section 31 of 36

31. Use PSLE Science to test explanations

PSLE Science develops the link between evidence and mechanism. Students can compare two candidate routes, decide which variable to change and predict what result would support or weaken each explanation. This is the same logic that makes a perturbation useful.

Contents · Previous section · Next section

Section 32 of 36

32. Connect Secondary and O-Level Science

Secondary Science and O-Level Biology develop cells, receptors, coordination, gene expression and experimental design. Domino provides a modern context for those ideas. The goal is not memorising software but learning to trace a claim through several evidence layers.

Contents · Previous section · Next section

Section 33 of 36

33. Let Computing strengthen reproducibility

Computing contributes matrices, correlation, graphs, version control and transparent pipelines. Biology supplies the molecular meaning and validation rules. A technically correct pipeline can still produce a biologically weak claim if versions, identifiers or specimens are mishandled.

Contents · Previous section · Next section

Section 34 of 36

34. Choose programmes by verified fit

When comparing science enrichment or school choices, use current official sources and look for sustained opportunities in biology, data analysis, computing, research and evidence evaluation. A programme name does not guarantee admission, mentoring, facilities or a career outcome.

Contents · Previous section · Next section

Section 35 of 36

35. See connected career families

The reasoning connects to bioinformatics, immunology, regenerative medicine, molecular biology, data science, research software and experimental design. Routes may pass through different education pathways. Requirements change, so learners should verify current course and employer information directly.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with the evidence direction intact

A defensible Domino result names the implementation, TF method, receptor association, ligand–receptor resource, specimens and sensitivity checks, then separates computational support from protein, spatial, temporal and perturbation validation. Working backwards from receiver activity is useful precisely because every link remains testable.

Domino is easiest to understand as an evidence chain that begins inside the receiver. A transcription-factor activity pattern supplies the downstream clue; receptor correlations suggest possible upstream entry points; a ligand–receptor resource supplies candidate source molecules. Each connection narrows the hypothesis space, and none should be hidden behind the final network diagram.

Did you know? The original Domino repository is archived, while dominoSignal is an actively developed successor. That is a reproducibility lesson in miniature: a paper, its preserved implementation and a current package can share an intellectual lineage without being interchangeable software versions.

Start an analysis with a release boundary. If reproducing the 2021 study, preserve the archived code, environment and knowledge resources. If using current dominoSignal, cite its current documentation and release, record defaults and describe the analysis as a contemporary implementation. This prevents later functionality from being projected backwards onto the publication.

The receiver-first direction can reduce a common problem in ligand–receptor analysis: a large expression-derived network with little evidence of cellular response. Yet TF activity itself is inferred from expression and regulons. Inspect the target genes that support a leading TF, determine whether support is distributed or dominated by one target, and compare specimens before using the score as a receiver-state anchor.

Correlation is a vulnerable bridge. Cell cycle, stress, treatment and cell subtype can raise both receptor expression and TF activity. Plot each candidate within specimens and relevant strata; consider mixed models or pseudobulk summaries that respect the study design. A pooled correlation that reverses or disappears inside specimens should not be presented as a robust receptor link.

Ligand mapping adds another versioned layer. Freeze the exact database, species, identifiers and complex rules. Show whether one receptor has many possible ligands and whether those ligands are expressed in anatomically plausible source cells. The most abundant partner is not automatically the active one.

Use a staged validation ladder. First, confirm source and receiver labels and replicate the component scores. Second, verify ligand and receptor protein in the proper cellular compartments. Third, measure a receptor-proximal event. Fourth, perturb the route and test the predicted TF programme. Each successful gate strengthens a different link; failure tells you precisely where the reconstruction breaks.

Negative results deserve space. A route may have stable TF activity but no receptor protein, a strong receptor correlation but no spatially plausible ligand source, or a database-compatible ligand with no downstream response. Reporting these cases teaches readers that inference is a process of elimination, not a machine for manufacturing certainty.

For learners, Domino links familiar habits across stages of education: careful observation in Primary Science, controlled comparisons in PSLE Science, cells and regulation in Secondary and O-Level Biology, and correlations, graphs and version control in Computing. The rewarding idea is that advanced software still depends on elementary honesty about what was measured.

The bounded conclusion is powerful enough: Domino can prioritise receiver-anchored ligand–receptor–TF hypotheses for targeted experiments. It does not directly observe ligand release, binding, receptor activation or causal information flow. Preserving that boundary makes every follow-up experiment clearer.

A practical report should include a compact evidence ledger for every headline chain: TF activity and regulon support, receptor correlation within each specimen, candidate ligands and source groups, spatial opportunity, protein evidence and perturbation status. Mark each field as observed, inferred, unsupported or not tested. This turns a network into an auditable research plan and prevents later readers from mistaking a missing experiment for a negative result.

Finally, compare Domino with a simpler expression-only baseline. If the receiver-first method adds value, it should nominate a more coherent and discriminating route than ligand–receptor abundance alone. When both approaches agree, that consensus is useful; when they disagree, the discrepancy is an opportunity to test whether downstream response evidence changes the biological priority.

Contents · Previous section · Continue to the Science Learning Hub