eduKateSG · Why Science?
Capture RNA from a tissue section on a dense field of two-micrometre barcoded beads, reconstruct fine spatial patterns, and keep decoding yield, molecule counts, diffusion, binning and specimen replication visible
Reading routes
Science learning becomes useful when a familiar object or observation is turned into a system of quantities, mechanisms and claim limits. This guide owns one applied evidence-reading job inside eduKateSG’s wider Science estate. It connects naturally to Why Science Spatial Transcriptomics Tissue Coordinates Gene Expression Evidence; Why Science Slide Seq Dna Barcoded Bead Pucks Spatial Transcriptome Evidence; Why Science Stereo Seq Dna Nanoball Patterned Array Mouse Organogenesis Spatial Transcriptome Evidence; Why Science Single Cell Rna Sequencing Barcodes Transcriptome Heterogeneity Evidence; Education Hub; Singapore Secondary School Directory; Career Adulthood Hub. It also keeps current school and public claims traceable to visible primary sources: Foundational HDST primary study; Foundational HDST PubMed record; Foundational HDST analysis code; 2026 Singapore–Cambridge O-Level Chemistry syllabus; 2026 Singapore–Cambridge O-Level Biology syllabus. The sources describe the scientific scope; this article translates that scope into a calm route for Primary Science, PSLE Science, Secondary Science, O-Level Science, STEM exploration, school choices and career pathways without inventing admission or employment outcomes.
High-definition spatial transcriptomics—HDST—captures RNA from histological tissue sections on a dense, spatially barcoded bead array. The 2019 Nature Methods study reported several hundred thousand transcript-coupled spatial barcodes per experiment at two-micrometre feature resolution, with demonstrations in mouse olfactory bulb and primary breast-cancer tissue. Fine coordinate spacing is powerful, yet effective biological resolution also depends on bead decoding, RNA capture, diffusion, molecule counts, tissue registration, binning, segmentation and independent specimens.
Inside this guide
1–12 · Foundations and models
- 1. Begin with coordinates smaller than most cells
- 2. Understand the two-micrometre statement
- 3. Give every bead an address
- 4. Separate spatial density from molecular sensitivity
- 5. Use histology as an independent layer
- 6. Remember the 2019 demonstrations
- 7. Define the question before choosing the bin
- 8. Build the bead pool deliberately
- 9. Decode across repeated optical cycles
- 10. Map decoded beads to sequence barcodes
- 11. Prepare a section that lies flat
- 12. Let RNA meet capture probes without erasing place
13–24 · Evidence, testing and applications
- 13. Use molecular identifiers to count copies
- 14. Register morphology and array coordinates
- 15. Practise with an invented HDST quality table
- 16. Inspect quality as a spatial map
- 17. Validate familiar anatomical markers first
- 18. Choose binning as an explicit model
- 19. Segment only where morphology supports it
- 20. Transfer cell labels with probabilities
- 21. Keep the specimen as the replicate
- 22. Distinguish pitch from effective resolution
- 23. Challenge low counts
- 24. Challenge lateral movement
25–36 · Learning, decisions and pathways
- 25. Challenge barcode clashes
- 26. Challenge manufacturing and day effects
- 27. Interpret the cancer example responsibly
- 28. Compare HDST with Slide-seq
- 29. Build a classroom bead-array model
- 30. Connect to Primary Science observation
- 31. Strengthen PSLE Science process skills
- 32. Extend into Secondary and O-Level Science
- 33. Use the topic to investigate school fit
- 34. See the career network behind the map
- 35. Did You Know? Smaller features may need larger groups
- 36. End with a bounded HDST conclusion
Section 1 of 36
1. Begin with coordinates smaller than most cells
HDST places a histological section over a dense lawn of spatially barcoded beads. Captured RNA molecules inherit bead coordinates, creating a fine-grained expression field. The coordinate pitch can be much smaller than a cell, but the biological meaning still depends on molecules recovered, tissue contact and registration.
Section 2 of 36
2. Understand the two-micrometre statement
The foundational study described transcript-coupled spatial barcodes at two-micrometre resolution. That number refers to array features and sampling coordinates. It should not be rewritten as guaranteed two-micrometre biological resolution, single-organelle truth or one perfectly measured cell per feature.
Section 3 of 36
3. Give every bead an address
Beads carry DNA barcodes assembled through split-and-pool chemistry. Optical decoding connects each bead’s sequence identity to an x–y position before or alongside tissue measurement. If identity and coordinate separate, later sequencing can place real transcripts in the wrong neighbourhood.
Section 4 of 36
4. Separate spatial density from molecular sensitivity
A dense array offers many possible sampling locations. Sensitivity asks how many RNA molecules and genes are actually captured at each one. Very fine features can be sparse, so neighbouring positions may need to be grouped for stable analysis.
Section 5 of 36
5. Use histology as an independent layer
A stained or bright-field image shows tissue shape, nuclei and morphological regions. The molecular map should be registered to this image rather than treated as self-locating. Histology can validate layers while also revealing tears, folds and empty areas.
Section 6 of 36
6. Remember the 2019 demonstrations
The Nature Methods paper demonstrated HDST in mouse olfactory bulb and primary breast-cancer tissue. Those examples establish feasibility in defined specimens. They do not prove equal performance in every organ, preservation method, disease or laboratory.
Section 7 of 36
7. Define the question before choosing the bin
A project may ask about a broad layer, a tumour boundary or a rare cellular niche. The scale of that question should set molecule-depth targets, binning, segmentation, section orientation and replicate count. Choosing scale after viewing the prettiest map invites overfitting.
Section 8 of 36
8. Build the bead pool deliberately
Barcode diversity must comfortably exceed the number of beads used. Sequence distance, synthesis quality and contamination influence collisions. Retain lot records and a barcode dictionary so unexpected hot spots can be checked against array manufacturing rather than assumed biological.
Section 9 of 36
9. Decode across repeated optical cycles
The study’s supplementary description used sequential hybridisations to decode bead identities. Each cycle contributes colour information. Focus drift, uneven illumination and missed colour calls can lower mapping yield, so confidence and rejected beads should be reported across the entire array.
Section 10 of 36
10. Map decoded beads to sequence barcodes
Optical patterns must be connected to the nucleotide barcodes later recovered in sequencing. Redundant or clashing assignments are filtered. Publish the proportion decoded, mapped, rejected and retained; a final map alone conceals where coordinate certainty was lost.
Section 11 of 36
11. Prepare a section that lies flat
Thickness, freezing, cutting and mounting influence RNA integrity and contact. Folds or partial adhesion create local capture deficits. Image the whole section before destructive processing and record its orientation so apparent tissue gaps can be compared with visible damage.
Section 12 of 36
12. Let RNA meet capture probes without erasing place
Permeabilisation releases transcripts toward nearby capture oligonucleotides. Time and chemistry must balance yield against lateral spread. A molecule’s assigned coordinate reflects the capture process, not an untouched snapshot of its original nanometre position.
Section 13 of 36
13. Use molecular identifiers to count copies
Unique molecular identifiers help distinguish captured molecules from amplification duplicates. Sequencing depth, saturation, valid barcodes, UMI totals and genes per feature describe technical performance. More reads cannot recover RNA that never reached a bead.
Section 14 of 36
14. Register morphology and array coordinates
Align tissue images to decoded bead positions with visible landmarks or fiducials. Report scale, rotation, residual error and any non-rigid correction. A narrow boundary claim becomes fragile when registration uncertainty approaches the boundary width.
Section 15 of 36
15. Practise with an invented HDST quality table
These fictional regions show why fine spacing alone cannot decide usability.
| Region | Decoded positions | Median UMIs per 10 μm bin | Registration error | First reading |
|---|---|---|---|---|
| A | 95% | 610 | 2.2 μm | strong starting region |
| B | 58% | 600 | 2.4 μm | coordinate gaps |
| C | 94% | 48 | 2.1 μm | weak molecular capture |
| D | 93% | 590 | 17.0 μm | boundary unreliable |
The values are invented for teaching, not study results.
Section 16 of 36
16. Inspect quality as a spatial map
Plot decoded-bead confidence, UMIs, genes, background and empty-feature rates at their coordinates. A failed patch can look like a biological void after smoothing. Keep the quality mask visible beside every expression map.
Section 17 of 36
17. Validate familiar anatomical markers first
Known olfactory-bulb layers or independently stained regions offer a pipeline check. Concordant marker patterns support capture, registration and analysis. Disagreement may indicate orientation, diffusion, sparse counts or annotation problems before it suggests unexpected biology.
Section 18 of 36
18. Choose binning as an explicit model
Grouping nearby two-micrometre features raises molecule counts while reducing spatial detail. Compare several declared bin sizes and show how boundaries and cell-type assignments change. A result that appears only after one convenient bin deserves cautious language.
Section 19 of 36
19. Segment only where morphology supports it
Nuclear outlines can guide aggregation toward cell-like units, yet cytoplasm, processes and extracellular RNA extend beyond nuclei. Report how features are assigned near borders and retain the original bead-level matrix for alternative analyses.
Section 20 of 36
20. Transfer cell labels with probabilities
Single-cell RNA-sequencing references can help interpret spatial profiles. Species, tissue state, sampling and batch must match closely enough. Present uncertain or mixed assignments instead of forcing every bin into a familiar cell type.
Section 21 of 36
21. Keep the specimen as the replicate
Features nest within bins, bins within sections and sections within people or animals. Hundreds of thousands of positions improve within-section description. They do not turn one specimen into a population study.
Section 22 of 36
22. Distinguish pitch from effective resolution
Feature spacing, RNA diffusion, registration, molecule counts and chosen aggregation all contribute to effective resolution. Report the scale supported by the complete chain. The smallest number printed in a methods paper is not automatically the smallest trustworthy biological claim.
Section 23 of 36
23. Challenge low counts
A silent position may reflect low abundance, degraded RNA, poor contact or limited capture. Estimate detection probability with positive controls and replicate sections. Reserve absence language for targets and regions where the assay had enough sensitivity to find a meaningful signal.
Section 24 of 36
24. Challenge lateral movement
Permeabilisation can blur a sharp interface. Test known boundaries, spike-in patterns or simulations and compare apparent width with histology. If diffusion is not measured, describe fine gradients as captured spatial patterns rather than exact molecular origins.
Section 25 of 36
25. Challenge barcode clashes
Two bead positions linked to the same or confused code can create duplicated or displaced expression. Quantify clashing barcodes and unexpected sequence combinations. Surprising isolated signals should survive barcode-confidence filtering before they become rare-cell stories.
Section 26 of 36
26. Challenge manufacturing and day effects
Bead packing, decoding, tissue handling and sequencing vary by array lot and processing date. Balance biological groups across these factors. A computational batch correction cannot fully recover evidence when all controls and cases were processed on different arrays.
Section 27 of 36
27. Interpret the cancer example responsibly
The breast-cancer demonstration showed that fine spatial measurements can distinguish tissue regions and cell-type enrichments. It did not establish diagnosis, prognosis or treatment choice. Human research claims require consent, independent cohorts and clinical validation beyond a technology demonstration.
Section 28 of 36
28. Compare HDST with Slide-seq
Both approaches connect transcript counts to barcoded beads. Slide-seq’s foundational puck used approximately cell-sized beads, while HDST pursued denser two-micrometre features followed by aggregation or segmentation. The useful comparison is sensitivity, effective resolution, field, workflow and question—not which acronym sounds newer.
Section 29 of 36
29. Build a classroom bead-array model
Students assign codes to tiny grid points, overlay a paper tissue and transfer coloured RNA tokens. They then group points into larger bins. The exercise makes the trade-off between fine coordinates and stable counts visible without biological samples.
Section 30 of 36
30. Connect to Primary Science observation
Young learners can compare where objects were placed with where coloured marks are later found. A blank coordinate may mean no object, no transfer or a failed detector. That simple distinction builds careful observation language.
Section 31 of 36
31. Strengthen PSLE Science process skills
Pupils can identify independent variables, controls and repeats in the invented table, calculate proportions and explain why registration matters. They practise separating a measurement from an inference rather than memorising specialist vocabulary.
Section 32 of 36
32. Extend into Secondary and O-Level Science
Biology contributes tissues, cells and gene expression; Chemistry contributes binding and reaction conditions; Mathematics contributes coordinates and aggregation; Computing contributes barcodes and classification. The method is enrichment anchored in scientific process and data interpretation.
Section 33 of 36
33. Use the topic to investigate school fit
Families can consult official school information for laboratory inquiry, data science, microscopy, research mentoring and ethics. Public spatial datasets allow meaningful analysis without a specialised array. Do not infer admissions, equipment access or guaranteed programmes.
Section 34 of 36
34. See the career network behind the map
HDST connects molecular biology, histology, bead chemistry, imaging, sequencing, statistics, software and clinical research governance. Each role needs current training and supervised practice. The article explains the collaboration without promising a qualification or job outcome.
Section 35 of 36
35. Did You Know? Smaller features may need larger groups
Two-micrometre coordinates can be combined into cell-like or region-like bins to gain counts. That is not failure; it is a transparent resolution–sensitivity choice. Good reporting shows both the fine measurements and the aggregation used for the claim.
Section 36 of 36
36. End with a bounded HDST conclusion
A defensible statement names tissue, arrays, decoded fraction, molecule depth, registration, effective scale, binning or segmentation and independent specimens. It may describe replicated spatial expression or enrichment. It should not turn fine feature spacing into perfect cellular truth, clinical prediction or causal biology.
Pre-registration can name the primary tissue region, expression features, minimum decoding yield, molecule-depth gate, registration ceiling, main bin or segmentation scale and independent specimen unit. Exploratory domains remain useful when labelled. This prevents repeated smoothing and aggregation until one visually dramatic boundary appears.
Array quality deserves its own release. Publish bead packing, optical-decoding confidence, barcode mapping, clashing-code removal, empty-feature rate and retained positions as spatial layers. Keep raw decoding images and the sequence dictionary. A later analyst should be able to distinguish a molecular pattern from a manufacturing pattern without relying on the final paper figure.
Batch monitoring should use a stable tissue or RNA reference where practical. Track decoded fraction, UMIs, genes, background, saturation, histology registration and effective boundary width by array lot, processing date and operator. Biological groups should be distributed across lots. If every control sits on one array batch and every case on another, no downstream correction can recreate the missing experimental comparison.
Binning should be handled like a measurement choice, not cosmetic rendering. Declare the primary bin before testing the biological hypothesis, then publish sensitivity at finer and coarser scales. Report how many molecules and cells contribute to each unit. Interpolation may improve readability, but the unsmoothed feature map and missing positions must remain visible.
Segmentation introduces a different model from square or circular binning. Nuclear masks can approximate cellular units, yet large cells, small cells and long processes violate a single rule. Compare mask-based and neighbourhood-based aggregation. Keep direct molecule counts, inferred cell identity and tissue-region annotation as separate evidence layers rather than merging them into one confident colour.
Spatial statistics must respect autocorrelation and hierarchy. Adjacent features share tissue environment and technical conditions, while thousands of genes create many candidate patterns. Predefined regions or corrected exploratory tests reduce false discovery. Summaries should show every independent section or specimen, not only a pooled feature-level significance value.
Negative results need a scale-specific detection boundary. Estimate the expression change and spatial-domain width supported by captured molecules, feature recovery, diffusion, registration and specimen number. A well-controlled null may exclude a broad, abundant domain while leaving rare cells or sharply localised low-abundance transcripts unresolved.
Protocol transfer requires fresh validation. Brain layers, fibrotic tissue, fatty tissue, plants and archived clinical material differ in sectioning, permeability, RNA quality and morphology. Re-establish contact, diffusion, capture and segmentation. A platform capable of two-micrometre features in one preparation does not guarantee the same effective performance elsewhere.
Human-tissue work requires consent, de-identification, secure sequence handling and limits on re-identification. Tumour maps can generate research hypotheses about cellular neighbourhoods, but they are not diagnostic or prognostic products. Any clinical claim would need independent cohorts, locked analysis, performance metrics and regulatory review.
A concise result might read: ‘Across independent mouse olfactory-bulb sections, quality-filtered HDST arrays recovered replicated layer-associated expression at a declared effective scale after barcode, registration and sensitivity controls.’ It should state specimens and uncertainty. It should not say that every two-micrometre feature was a directly observed cell or organelle.
Reproducibility also depends on complete data stewardship. Retain raw images, bead decoding, barcode dictionaries, histology, registration transforms, sequencing files, UMI rules, quality masks, binning and segmentation parameters, references, code, versions and checksums. Record failed arrays and excluded regions with reasons. The next experiment should target the leading uncertainty rather than simply generating a larger colour map.
Resolution claims become clearer when authors publish a scale ledger. For every figure, name bead pitch, registration residual, diffusion estimate, chosen bin, segmentation rule and the smallest interpreted domain. Readers can then see which scale comes from hardware, which from computation and which from biology. This small table also prevents a two-micrometre platform label from being copied into a conclusion whose evidence actually rests on much larger aggregated regions.
Precise verbs keep the record honest: measured, mapped, associated, inferred, predicted or experimentally changed are not interchangeable. Laboratory work requires approved biological handling, chemical risk controls, trained instrument use and lawful waste disposal; classroom work should use synthetic or public data. Reports should publish exclusions, failed runs, denominators, effect sizes and uncertainty beside the successful map.
Claims should travel with their scale—molecule, feature, pixel, cell, field, section, specimen or population. Moving between these levels without saying so is a common source of false certainty.
Contents · Previous section · Continue to the Science Learning Hub
