Why Science? | SpaTalk, Spatial Ligand–Receptor–Target Networks and Knowledge-Graph Evidence

Section 1 of 36

1. Start with two linked components

SpaTalk can reconstruct cell-level spatial information from spot-based data and then infer communication through spatial and molecular knowledge. These components answer different questions. Reconstruction estimates where reference-like cells may sit; communication scoring evaluates candidate ligand–receptor–target routes.

Contents · Next section

Section 2 of 36

2. Keep the 2022 study boundary

SpaTalk was published in Nature Communications in 2022 and evaluated across spatial technologies and applications. Those demonstrations establish feasibility for the reported data. They do not guarantee the same accuracy in every tissue, reference atlas or spatial resolution.

Contents · Previous section · Next section

Section 3 of 36

3. Decompose mixed spots

For spot-based spatial transcriptomics, the workflow estimates cell-type composition using a non-negative linear model. A spot can contain several cells, so composition is uncertain. Report fit, reference coverage and ambiguous cell types rather than treating proportions as direct counts.

Contents · Previous section · Next section

Section 4 of 36

4. Map reference cells into space

Reference single-cell profiles can be assigned to positions within spots to create a reconstructed cell-level view. These positions are computational placements constrained by the data and model. They should never be described as microscope-tracked cells unless independently validated.

Contents · Previous section · Next section

Section 5 of 36

5. Build a spatial neighbour graph

Candidate sender and receiver cells are linked through spatial neighbourhoods, commonly using a chosen neighbour rule such as K nearest neighbours. The choice of K and coordinate scale defines opportunity. A different radius or graph can change which routes are eligible.

Contents · Previous section · Next section

Section 6 of 36

6. Use the ligand–receptor–target graph

SpaTalk combines ligand–receptor knowledge with intracellular pathways and transcription-factor targets. This LRT graph lets a candidate extracellular interaction connect to a possible receiver programme. Every edge is prior knowledge, not proof that the path was active in the specimen.

Contents · Previous section · Next section

Section 7 of 36

7. Separate intercellular and intracellular evidence

The method scores evidence between source and receiver and evidence from receptor towards target genes. Publish both components. A high combined route can be driven by strong proximity and expression while the downstream path is weak, or the reverse.

Contents · Previous section · Next section

Section 8 of 36

8. Create a spatial manifest

Record platform, image or coordinate system, spot size, tissue section, specimen, preprocessing, reference dataset, deconvolution settings, neighbour graph, LRI and pathway resources, random seeds and SpaTalk version. Spatial claims are uninterpretable without geometry and resolution.

Contents · Previous section · Next section

Section 9 of 36

9. Audit the reference atlas

Check that the single-cell reference matches species, tissue, condition and cell states. Missing or mismatched cell types can force spots into the closest available category. Use marker review and alternative references for important reconstructed populations.

Contents · Previous section · Next section

Section 10 of 36

10. Protect coordinate integrity

Confirm orientation, units, tissue masks, removed spots and transformations before building neighbours. A flipped, scaled or misregistered coordinate table can produce an elegant but false neighbourhood network. Archive original and processed coordinates.

Contents · Previous section · Next section

Section 11 of 36

11. Freeze molecular knowledge resources

Save the ligand–receptor pairs, pathways, receptor-to-TF links and TF targets used. The published workflow draws on established resources and also allows custom inputs. Record every addition so that a novel edge is not mistaken for a default SpaTalk relationship.

Contents · Previous section · Next section

Section 12 of 36

12. Archive reconstruction and scoring

Retain cell-type proportions, mapped cells, coordinates, neighbour graph, permutations, random-walk outputs, intercellular scores, intracellular scores and complete LRT paths. A final chord diagram cannot reveal the assumptions that created it.

Contents · Previous section · Next section

Section 13 of 36

13. Preserve biological replication

Sections, spots and reconstructed cells are nested within specimens. Compare route support across independent people, animals or samples. Many spatial locations from one tissue improve mapping but do not create independent biological replication.

Contents · Previous section · Next section

Section 14 of 36

14. Plan permutation tests carefully

A spatial null should preserve relevant tissue structure and abundance while disrupting the claimed sender–receiver relationship. Randomising across incompatible anatomical regions can be too easy to beat. Explain what the permutation represents biologically.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented LRT table

This fictional table is for interpretation practice, not SpaTalk output. Each route has a different weakest layer. Choose validation that tests that layer.

Contents · Previous section · Next section

Section 16 of 36

16. Read reconstructed cells conditionally

Visualise the original spots, estimated proportions and mapped cells together. Report uncertainty or alternative placements. A reconstructed map is most useful as a testable spatial hypothesis, especially when several similar cell states compete for the same location.

Contents · Previous section · Next section

Section 17 of 36

17. Read proximal ligand–receptor evidence

Check source ligand expression, receiver receptor expression, neighbour distance and specimen prevalence. Confirm whether the mediator is contact-dependent, secreted or matrix-associated. The spatial rule should reflect that biology rather than a single generic KNN graph.

Contents · Previous section · Next section

Section 18 of 36

18. Trace downstream target routes

Display the receptor, intermediate pathway nodes, transcription factors and target genes. Check expression and provenance for important intermediates. A random walk can rank plausible paths through the graph; it does not observe biochemical traversal.

Contents · Previous section · Next section

Section 19 of 36

19. Compare specimens and sections

Estimate the route separately in each specimen and, when available, across serial or matched sections. A route that appears only where tissue quality or cell reconstruction differs should be reported as section-sensitive.

Contents · Previous section · Next section

Section 20 of 36

20. Audit neighbour scale

Repeat important findings across biologically reasonable K values or distance radii. A contact signal should not depend on distant neighbours, while a diffusible factor may. Report the range over which the conclusion is stable.

Contents · Previous section · Next section

Section 21 of 36

21. Audit deconvolution sensitivity

Compare reference choices, marker sets or alternative composition estimates. Track whether sender and receiver identities, locations and route ranks persist. A communication claim cannot be more certain than the reconstructed cell map that supplies its nodes.

Contents · Previous section · Next section

Section 22 of 36

22. Audit the random walk and permutations

Vary planned restart, weighting and significance settings while retaining the biological question. Publish component scores and adjusted values. A route that survives only one convenient setting should be labelled exploratory.

Contents · Previous section · Next section

Section 23 of 36

23. Check spatial resolution limits

STARmap, Slide-seq, Visium and other platforms capture different spatial scales and molecular breadth. Do not compare a cell-level coordinate and a multicellular spot as if they provide identical evidence. State what the platform can resolve.

Contents · Previous section · Next section

Section 24 of 36

24. Compare custom and default knowledge

SpaTalk allows custom ligand–receptor, pathway and TF information. That flexibility is valuable, but custom resources change the hypothesis universe. Separate default, curated additions and exploratory edges in the results.

Contents · Previous section · Next section

Section 25 of 36

25. Measure proteins in place

Use spatial protein or imaging assays to confirm ligand and receptor localisation in the nominated cells. Then measure a proximal receiver event and downstream target response. RNA proximity alone cannot establish binding or pathway activation.

Contents · Previous section · Next section

Section 26 of 36

26. Use time when routes imply order

A spatial snapshot can align source and receiver but not establish what happened first. Time courses after stimulation or perturbation can test whether receptor-proximal events precede transcription-factor and target changes in the receiver.

Contents · Previous section · Next section

Section 27 of 36

27. Design a spatially discriminating perturbation

Block a leading ligand or receptor and measure the predicted response in nearby and distant receiver cells. A distance-dependent change can distinguish local signalling from a global treatment effect. Include a route that should remain unchanged.

Contents · Previous section · Next section

Section 28 of 36

28. Write a bounded SpaTalk claim

Prefer: ‘SpaTalk prioritised a spatially proximal LRT route whose component scores were stable across neighbour settings and specimens.’ Avoid: ‘SpaTalk showed that one reconstructed cell signalled to another.’ The first is accurate and experimentally useful.

Contents · Previous section · Next section

Section 29 of 36

29. Map the workflow in class

Students can place cell cards on a tissue grid, draw neighbour rules, add ligand–receptor cards and trace downstream targets. Changing the radius or one cell placement shows how spatial assumptions alter a network without changing the molecular database.

Contents · Previous section · Next section

Section 30 of 36

30. Begin with Primary Science observation

Location, grouping and careful comparison are familiar scientific habits. Learners can observe which objects are near, define a fair distance rule and record exceptions. That same clarity is essential in spatial omics.

Contents · Previous section · Next section

Section 31 of 36

31. Use PSLE Science for variables

Students can identify the changed variable—neighbour radius, reference cells or ligand presence—and predict its effect. They learn that a model result depends on controlled choices and that a fair test changes one important factor at a time.

Contents · Previous section · Next section

Section 32 of 36

32. Connect Secondary and O-Level Science

Cells, transport, receptors, coordination and experimental design provide the biological language. SpaTalk turns those ideas into a spatial evidence problem, encouraging learners to distinguish proximity, expression and response.

Contents · Previous section · Next section

Section 33 of 36

33. Let Computing handle graphs

Computing contributes coordinates, nearest-neighbour graphs, random walks, permutations and reproducible code. Biology determines whether an edge is physically plausible. A graph algorithm cannot choose the correct biological distance without domain knowledge.

Contents · Previous section · Next section

Section 34 of 36

34. Choose programmes by current evidence

When considering STEM enrichment or school choices, verify official current information and look for authentic opportunities in biology, microscopy, data analysis, computing and research. Never infer admission or outcomes from a programme title.

Contents · Previous section · Next section

Section 35 of 36

35. See connected careers

Spatial analysis connects to pathology, developmental biology, bioinformatics, microscopy, data engineering, drug discovery and research software. Pathways differ and entry requirements evolve, so current official course and employer sources remain essential.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with place and mechanism separate

A defensible SpaTalk conclusion states the platform, reference, reconstruction, neighbour graph, LRT resources, specimens and sensitivity checks, then separates spatial opportunity from molecular activity and causal validation. The map guides experiments; it does not replace them.

SpaTalk joins two difficult tasks: reconstructing cellular composition and position from spot-based spatial transcriptomics, and linking nearby ligand–receptor evidence to possible downstream targets. The combination is attractive because it connects place with response. It also means uncertainty from reconstruction can propagate into every later communication score.

Did you know? A 10x Visium spot is not automatically a single cell. When several cells contribute RNA to one spot, a cell-level map is reconstructed from a reference and model. The plotted cell coordinates are therefore hypotheses constrained by spatial data, not direct observations of individual cell centres.

Begin with geometry. Preserve the original tissue image, spot coordinates, scale, masks and transformations. Verify orientation before any neighbour graph is built. Then define a biological neighbourhood: contact-dependent molecules need a tighter opportunity rule than diffusible factors. Sensitivity across reasonable K values or radii should be reported, not hidden.

Next, examine the reference. Missing cell states, different disease conditions or batch effects can push a spot towards the nearest available profile. Compare marker expression in the original spots with the reconstructed identities and test an alternative reference when a headline route depends on a rare or ambiguous population.

The ligand–receptor–target knowledge graph adds mechanistic context. Preserve the route from ligand to receptor, pathway intermediates, transcription factor and targets, with edge provenance. Random-walk or path scores prioritise connectivity in that graph; they do not show phosphorylation, nuclear entry or transcription occurring in the tissue section.

Keep intercellular and intracellular scores separate. A candidate can have excellent spatial ligand–receptor evidence but weak downstream support, or a compelling target programme with uncertain spatial source. Those two cases call for different experiments. A single combined rank should never conceal which side is carrying the result.

Validation should retain location. Confirm ligand and receptor proteins with imaging or spatial proteomics, then assay a proximal receiver event. If possible, perturb the source route and compare nearby with more distant receivers. A response that changes with the predicted neighbourhood is more discriminating than a whole-tissue average.

Platform differences matter. STARmap, Slide-seq, Visium, MERFISH and seqFISH+ differ in resolution, field, sensitivity and gene coverage. A method can accommodate several technologies without making their evidence equivalent. State which spatial claims the chosen platform can actually resolve.

Archive reconstructed maps, neighbour graphs and complete LRT paths alongside the original data. A future reference atlas or pathway update may move cells or rerank routes. Versioned objects let readers see that the analytical model changed; the historical tissue section did not.

For learners, SpaTalk offers a vivid bridge from maps and distances to cells and signals. Primary and PSLE Science supply observation and fair testing; Secondary and O-Level Science add transport and coordination; Computing adds coordinates, graphs and random walks. The same question persists: what does the map show, and what remains inferred?

The bounded conclusion is that SpaTalk can prioritise spatially plausible ligand–receptor–target routes from single-cell or spot-based spatial data. Reconstructed cells, molecular paths and scores remain model outputs. Direct localisation, protein activity, timing and perturbation supply the next evidence.

Quantify reconstruction uncertainty where possible. Compare alternative cell placements, composition estimates or references and record how often a proposed sender–receiver neighbourhood persists. A stable route should not depend on one arbitrary assignment of similar reference cells inside a mixed spot. When uncertainty is high, report the result at the cell-type or spot level rather than displaying false single-cell precision.

Use anatomy as a real control. Identify compartments where a route is physically plausible and compartments where it should not occur. A method that ranks the interaction equally in both places may be responding to broad expression rather than meaningful spatial opportunity. This is often more informative than a completely random coordinate shuffle.

For each headline LRT route, publish the path length, edge provenance, spatial distance, component scores and specimen prevalence. Long knowledge-graph paths can connect many molecules by chance, while short paths may be dominated by familiar canonical biology. Transparent path detail lets readers judge both possibilities.

Also distinguish discovery from confirmation. Use one section or subset to nominate routes, then test the fixed candidates in held-out specimens or sections. This reduces the temptation to choose the neighbour scale, reference or pathway resource after seeing the most appealing picture.

Treat a reconstructed absence cautiously. If a cell type is not placed near the proposed source, ask whether the reference contained that state and whether the platform could detect its markers. A true anatomical absence, a failed deconvolution and insufficient gene coverage require different conclusions and different follow-up work.

A balanced presentation should include one route supported by spatial and intracellular evidence, one proximity-only route, one knowledge-graph-sensitive route and one anatomical negative control. Showing disagreement makes the two-component design understandable and prevents a combined score from appearing magical.

Finally, retain tissue context in every export. Coordinates without section identity, orientation or scale can be reused incorrectly. Include a small provenance record with the specimen, platform, image transformation and neighbour definition so another analyst can reproduce the geometry before reproducing the network.

Contents · Previous section · Continue to the Science Learning Hub