Why Science? | MERINGUE, Spatial Autocorrelation and Nonuniform Cell-Density Evidence

Section 1 of 36

1. Begin with unevenly spaced cells

Spatially resolved assays often place cells irregularly rather than on a perfect grid. MERINGUE asks which genes and cell relationships show structure after accounting for this geometry. A pattern may be meaningful only if the neighbourhood definition and density variation are made explicit.

Contents · Next section

Section 2 of 36

2. Use spatial autocorrelation as the first lens

Spatial autocorrelation asks whether nearby observations tend to carry similar expression values. Positive local structure can reveal gradients, patches or layers. It does not explain why the pattern exists, and it can also reflect imaging, capture efficiency or tissue damage.

Contents · Previous section · Next section

Section 3 of 36

3. Construct a neighbourhood graph

Coordinates are converted into a graph that specifies which cells count as neighbours. MERINGUE uses this graph to calculate spatial statistics without assuming a uniform lattice. Every subsequent result inherits the graph’s choices, missing regions and boundary behaviour.

Contents · Previous section · Next section

Section 4 of 36

4. Adapt to nonuniform density

Dense and sparse regions should not be treated as if equal numerical distances imply equal biological context. The framework is designed to work in a density-agnostic manner, yet analysts must still inspect physical neighbour distances and sampling coverage rather than rely on the label alone.

Contents · Previous section · Next section

Section 5 of 36

5. Find spatially heterogeneous genes

Genes with expression arranged in tissue can show stronger spatial autocorrelation than expected under a suitable null. Such genes are candidates for follow-up. Statistical significance does not by itself establish a functional compartment or mechanism.

Contents · Previous section · Next section

Section 6 of 36

6. Use cross-correlation for paired patterns

Spatial cross-correlation asks whether one feature at a location relates to another feature nearby. This can nominate putative cell–cell relationships or coordinated processes. Shared cell type, anatomy and technical gradients remain plausible alternatives.

Contents · Previous section · Next section

Section 7 of 36

7. Keep the paper's demonstrated scope specific

The primary study applied MERINGUE to multiple spatially resolved datasets, including image- and sequencing-based assays, and supported analysis in two and three dimensions. These demonstrations establish capability under reported conditions, not automatic performance for every panel, tissue or resolution.

Contents · Previous section · Next section

Section 8 of 36

8. Lock expression rows to coordinate rows

Create a one-to-one audit between cell identifiers, expression columns and coordinates. Check duplicates, missing cells, axis direction, coordinate units and field joins. A reordered matrix can generate convincing but false spatial structure because the statistic cannot detect ownership errors.

Contents · Previous section · Next section

Section 9 of 36

9. Map quality before spatial statistics

Plot total counts, detected genes, background, segmentation confidence, tissue edges and acquisition fields. Spatially organised quality variation can be more autocorrelated than biology. Keep these maps beside every headline gene pattern.

Contents · Previous section · Next section

Section 10 of 36

10. Filter and normalise without erasing location

Document count filters, transformations, scaling and feature selection. Removing low-quality cells can change the graph; aggressive normalisation can change correlation. Repeat key findings with one defensible preprocessing alternative and preserve raw values for validation.

Contents · Previous section · Next section

Section 11 of 36

11. Inspect physical neighbour distances

Summarise how far graph edges travel in dense, sparse, interior and boundary regions. Neighbours that cross holes, tears or field gaps should be blocked. A graph can be topologically tidy while being physically implausible.

Contents · Previous section · Next section

Section 12 of 36

12. Define the spatial null clearly

A p-value needs a null model that respects the question. Coordinate permutation, expression permutation and graph rewiring remove different structures. State what is held fixed and what is broken so readers know which alternative the test addresses.

Contents · Previous section · Next section

Section 13 of 36

13. Control the many-gene search

Testing thousands of genes produces expected false positives. Report the number tested, filtering criteria and multiple-testing correction. Do not select only familiar markers after seeing the map and then present them as a prespecified set.

Contents · Previous section · Next section

Section 14 of 36

14. Archive graph and statistic settings

Save neighbour edges, weights, distance rules, statistic formulas, permutation count, random seeds, software version and coordinate transforms. Reproducibility requires more than a final list of significant genes.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented spatial audit table

This fictional table teaches interpretation, not performance.

Contents · Previous section · Next section

Section 16 of 36

16. Read a spatial gene as a hypothesis

A significant gene map proposes that expression is locally organised beyond the chosen null. Examine effect size, spatial scale, raw counts and cell composition. A tiny but significant statistic in a large dataset may have little biological consequence.

Contents · Previous section · Next section

Section 17 of 36

17. Separate cell type from within-type state

Spatial expression may change because different cell types occupy different locations. Repeat selected analysis within reliable cell types or adjust composition before claiming state regulation. Tissue composition and cell-intrinsic change are different owners.

Contents · Previous section · Next section

Section 18 of 36

18. Treat cross-correlation as association

A ligand-like gene near a receptor-like gene can motivate a signalling hypothesis, but transcriptional proximity is not molecule transfer. Direction, protein availability, receptor activation and timing require separate measurements.

Contents · Previous section · Next section

Section 19 of 36

19. Use spatial clustering carefully

Spatially informed features can group cells or regions, but cluster number and boundaries depend on the graph and algorithm. Show marker evidence, stability and morphology. Avoid converting soft gradients into crisp anatomical facts.

Contents · Previous section · Next section

Section 20 of 36

20. Extend to three dimensions with registration limits

Three-dimensional coordinates can reveal structures missed in a single section. If slices were registered computationally, registration error becomes part of the spatial statistic. Report slice spacing, transforms and uncertainty rather than treating the reconstructed volume as directly measured continuity.

Contents · Previous section · Next section

Section 21 of 36

21. Replicate at the specimen level

Thousands of cells from one tissue are not thousands of independent organisms. Show patterns across donors, animals or independently processed sections. Report consistent structure and genuine biological heterogeneity separately.

Contents · Previous section · Next section

Section 22 of 36

22. Audit graph construction sensitivity

Rebuild representative graphs with plausible neighbour or distance settings. Track leading genes, effect sizes and clusters. A conclusion that survives nearby choices is stronger than one tied to a single edge rule.

Contents · Previous section · Next section

Section 23 of 36

23. Audit density and boundary sensitivity

Subsample dense areas, remove edge bands and repeat the analysis. If a pattern follows acquisition density or field limits, call it sensitive. Density correction cannot replace direct inspection of sampling geometry.

Contents · Previous section · Next section

Section 24 of 36

24. Audit null-model calibration

Simulate unstructured expression over the observed coordinates and structured gradients of known scale. Check false-positive control and detection. Null calibration matters because tissue graphs contain irregular degrees and boundaries that can distort textbook assumptions.

Contents · Previous section · Next section

Section 25 of 36

25. Validate on raw and orthogonal measurements

Return to raw molecule counts, images, in-situ probes or protein staining for stable, weak and negative examples. Validation should not be another transformation of the same matrix alone.

Contents · Previous section · Next section

Section 26 of 36

26. Compare with a non-spatial baseline

Test whether ordinary differential variability or clustering recovers the same signal under matched preprocessing. MERINGUE earns interpretation when spatial information adds reproducible structure rather than merely changing the picture.

Contents · Previous section · Next section

Section 27 of 36

27. Report non-detection with a power envelope

Failure to detect autocorrelation may reflect sparse genes, a pattern smaller than the sampling interval, missing cells or genuine absence. Simulations on the observed geometry show what the experiment could have found.

Contents · Previous section · Next section

Section 28 of 36

28. Turn a pattern into a decisive measurement

Choose a candidate gene, region and predicted spatial scale, then design an independent assay or perturbation that can confirm or refute the arrangement. The map becomes useful when it narrows the next experiment.

Contents · Previous section · Next section

Section 29 of 36

29. Build a classroom model before code

Draw an irregular set of points, assign one broad gradient and one random value, then compare neighbour similarity before and after shuffling the values. Keep measured values, derived quantities, assumptions and conclusions in separate columns. The goal is not to imitate specialist software; it is to make the evidence chain visible enough to question.

Contents · Previous section · Next section

Section 30 of 36

30. Start with Primary Science habits

Primary Science already teaches the habits beneath MERINGUE: observe carefully, compare fairly, record consistently and avoid a conclusion larger than the test. A simple map of plant height, shade or water can show that location changes interpretation without introducing advanced statistics.

Contents · Previous section · Next section

Section 31 of 36

31. Use PSLE Science to practise claim limits

PSLE Science rewards links between observation, process and explanation. MERINGUE offers a modern reminder that a pattern is not automatically a cause. Students can ask what was changed, what was measured, what stayed uncontrolled and what extra observation would distinguish competing 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, variation, experimental design and evaluation. MERINGUE adds a contemporary case where molecular measurements and tissue position meet. The useful learning outcome is controlled reasoning, not memorising a package name.

Contents · Previous section · Next section

Section 33 of 36

33. Let Computing support the science

Computing contributes data structures, algorithms, visualisation, optimisation and reproducibility. Biology supplies the specimen, mechanism and validation. MERINGUE shows why reliable code is necessary but insufficient: an algorithm can execute perfectly on mismatched coordinates or a biased sample.

Contents · Previous section · Next section

Section 34 of 36

34. Choose school opportunities by fit

Families should verify laboratory practice, Biology, Computing, research programmes, CCAs and Applied Learning opportunities on current official school and MOE pages. A school mention of genomics, AI or data science does not guarantee depth, admission or a career outcome. Look for sustained inquiry, careful teaching, accessible mentoring and time to explain evidence.

Contents · Previous section · Next section

Section 35 of 36

35. See the connected career families

Reasoning used in MERINGUE appears in spatial biology, bioinformatics, tissue imaging, statistics, pathology and scientific software. Routes can pass through polytechnic, junior college, university or continuing education with different blends of biology, mathematics, statistics, computing and communication. Current course requirements must be checked directly; one project cannot guarantee entry or employment.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a bounded scientific claim

A defensible conclusion says exactly what MERINGUE estimated and under which inputs, settings and specimens. MERINGUE estimates spatial association on a chosen graph; it does not turn correlation into signalling, clusters into anatomy or registered slices into continuous tissue. Report uncertainty, sensitivity, replication, negative results and orthogonal validation together. That boundary is not weakness; it is what makes the result testable and reusable.

A publication-ready MERINGUE workflow begins with a spatial data contract. Record assay, specimen, expression layer, segmentation or spot unit, coordinate system, physical units, field joins, tissue mask, excluded observations and the identifier used to link every profile to its location. Preserve raw coordinates and the graph-ready coordinates separately so a rotation, crop or rescaling can be audited.

Write the neighbourhood rule before looking at the most attractive gene maps. State whether the graph is based on Delaunay relationships, k-nearest neighbours, a radius or another rule; show degree and edge-length distributions; and explain how holes, disconnected components and tissue borders are handled. Uneven density is not a nuisance that disappears because the software is density-aware.

Pre-register the primary statistical task. Finding spatially heterogeneous genes, studying cross-correlation, clustering cells and reconstructing three-dimensional patterns require different nulls and validation. One significant list should not quietly become evidence for every downstream story.

Map technical quality on the same axes as expression. Depth, detected features, background, segmentation area, illumination, section, batch and field boundaries can all be spatially organised. When a quality variable is autocorrelated, test whether headline genes remain structured after sensible adjustment or stratification.

Multiple testing belongs in the visible results. Report tested genes, excluded genes, raw and adjusted significance, effect size and spatial scale. A massive cell count can make weak effects statistically convincing. Readers need enough information to distinguish reproducible organisation from numerical sensitivity.

Use several null checks because each removes different information. Permuting expression across fixed coordinates tests location dependence; rewiring edges tests specific adjacency; rotating or translating fields tests acquisition-linked structure; simulated gradients test detection. Describe the null question beside its result.

Cross-correlation requires additional restraint. Shared anatomy can make two genes locally associated even when their products never interact. Evaluate within cell types or domains, adjust major composition where appropriate and demand protein, perturbation or temporal evidence before using mechanistic language.

For spatial clustering, publish a stability grid over graph and cluster settings. Show raw markers, spatial composition, morphology and uncertain boundaries. If known anatomy sets cluster number, call the exercise label-informed benchmarking rather than independent discovery.

Three-dimensional analysis must expose registration. Show every measured plane, transform, spacing and misalignment estimate. Do not interpolate through unmeasured gaps or across missing tissue without clear visual distinction. A 3D point cloud can be useful while remaining a reconstructed coordinate system.

Reproducibility materials should include matrices, coordinates, masks, graph edges and weights, expression transformations, statistics, null-generation code, seeds, multiple-testing results, clusters, environment, software commit, figures and checksums. Archive the full result table, not only familiar genes.

Did you know? Two datasets can show the same biological pattern yet produce different nearest-neighbour graphs when one part of the tissue is sampled more densely. That is why density, geometry and boundary checks are not cosmetic extras; they help decide whether a spatial statistic represents biology or the way observations were collected.

A careful final sentence might say: ‘MERINGUE identified spatial expression and neighbourhood associations that were stable across prespecified graph, preprocessing and null choices, repeated in independent specimens and aligned with orthogonal tissue evidence.’ It should not call autocorrelation mechanism, cross-correlation signalling or a registered volume directly measured continuity.

One useful teaching exercise is to give students three synthetic tissues: a smooth gradient, a sharp two-domain boundary and random expression sampled at uneven density. Ask them to predict which pattern a global autocorrelation statistic will emphasise, then inspect local maps and graph edges. The exercise makes a durable point: a summary statistic can identify spatial organisation without describing its shape, scale or cause.

Border effects deserve their own figure. Cells near a tissue edge have fewer possible neighbours, while cells beside a tear or cavity may be close in an image yet biologically separated. Recalculate representative results after trimming a narrow border, blocking edges across masks and changing the neighbourhood radius. Agreement across these checks makes the conclusion easier to trust; disagreement identifies where the tissue geometry is doing real analytical work.

For cross-gene analyses, compare observed local associations with associations expected from cell-type composition. A pair of markers may co-vary because the same cell population occupies a region, because two populations meet at a boundary, or because regulation connects them. These explanations suggest different follow-up experiments. Spatial statistics can rank them, but microscopy, perturbation and time-resolved evidence are what separate proximity from mechanism.

Finally, turn the analysis into an evidence ladder. Level one is a reproducible spatial statistic; level two is the same pattern across specimens; level three is agreement with histology or protein imaging; level four is a perturbation that changes the predicted spatial relationship. This ladder keeps an exciting map connected to the strength of the evidence and shows students how scientific confidence grows step by step.

Contents · Previous section · Continue to the Science Learning Hub