Why Science? | sci-Space, Combinatorial Indexing and Single-Cell Spatial-Transcript Evidence

Section 1 of 36

1. Start with a spatial hash, not a photograph

sci-Space assigns nuclei an approximate address by transferring known hashing oligonucleotides from a patterned slide. Later sequencing recovers the hash beside each nucleus’s transcriptome. The method joins single-cell molecular depth with a coarse spatial label rather than keeping every nucleus visibly fixed during measurement.

Contents · Next section

Section 2 of 36

2. Understand why dissociation usually loses place

Single-cell RNA sequencing separates cells or nuclei to profile them individually, which normally destroys tissue coordinates. sci-Space writes location information onto nuclei before extraction. This preserves a region code through dissociation, but the code inherits the geometry and diffusion limits of the patterned hash.

Contents · Previous section · Next section

Section 3 of 36

3. Pattern unique oligos across a large area

The study arrayed combinations of hashing oligos on an agarose-coated slide. Each spot represented a known region. Scale came from covering whole embryo sections, while location precision remained tied to spot radius, spacing and transfer rather than the dimensions of an individual nucleus.

Contents · Previous section · Next section

Section 4 of 36

4. Transfer hashes by controlled diffusion

A permeabilised tissue section is juxtaposed with the patterned slide so hashing oligos move into nearby nuclei. Diffusion makes the label possible and creates uncertainty. Concentration, time, contact and tissue structure affect how sharply one region can be separated from the next.

Contents · Previous section · Next section

Section 5 of 36

5. Add a section identity before pooling

Slide-specific oligos allow several sections to be pooled while retaining their origin. Pooling improves workflow efficiency, but sample swaps or index leakage can misassign entire groups. Section identity and local spatial hash should be audited as separate levels of information.

Contents · Previous section · Next section

Section 6 of 36

6. Profile nuclei with combinatorial indexing

After transfer, nuclei are extracted and processed by sci-RNA-seq. Repeated indexing steps create single-nucleus transcriptome labels without isolating every nucleus in its own well. Collision and doublet detection remain essential because two nuclei sharing an index combination can resemble one unusual cell.

Contents · Previous section · Next section

Section 7 of 36

7. Keep the published scale exact

The foundational report analysed fourteen sagittal sections from two E14.0 mouse embryos and retained 121,909 spatially resolved single-cell transcriptomes. Those are impressive cell counts but only two biological embryos. Cell number strengthens within-sample description; it does not replace animal-level replication.

Contents · Previous section · Next section

Section 8 of 36

8. Register the hash grid to the tissue image

Fluorescent fiducials help align patterned spots with the embryo section. Registration should be checked locally because the assembly may rotate, stretch or shift. A correct hash sequence can still point to the wrong anatomical region if image alignment fails.

Contents · Previous section · Next section

Section 9 of 36

9. Measure spot shape and spacing

Nominal grid design does not guarantee identical deposited spots. Report radius, centre spacing, missing spots and overlap. Region assignments near spot edges deserve lower confidence, especially when diffusion expands the effective footprint beyond the printed geometry.

Contents · Previous section · Next section

Section 10 of 36

10. Optimise permeabilisation for nuclei and RNA

The tissue must admit spatial oligos while preserving nuclei and transcriptomes. Too little transfer lowers hash detection; too much treatment can increase diffusion or RNA loss. Pilot experiments should evaluate hash specificity and transcriptome quality together rather than optimising either in isolation.

Contents · Previous section · Next section

Section 11 of 36

11. Recover nuclei without favouring easy tissues

Dissociation can release some cell types more readily than others. Compare recovered cell composition with histology or reference data and report debris and loss. Spatial patterns among recovered nuclei do not automatically represent every cell present in the original section.

Contents · Previous section · Next section

Section 12 of 36

12. Detect and remove barcode collisions

Combinatorial indexing assigns cells through multiple barcode rounds, yet collisions remain possible. Use species mixtures, scrublet-like models or other justified controls. Removing probable doublets is useful, but aggressive filters can also remove true transitional or high-RNA cell states.

Contents · Previous section · Next section

Section 13 of 36

13. Map cell types with references transparently

The original analysis integrated sci-Space profiles with non-spatial embryo atlases to annotate cell types. Reference mapping adds biological names; it is not direct spatial measurement. Report dataset version, age match, label probabilities and unresolved cells so inference remains visible.

Contents · Previous section · Next section

Section 14 of 36

14. Treat approximate coordinates honestly

A spatial hash can locate a cell to a patterned region, not necessarily to its exact original centre. State the region dimensions and assignment confidence in figures. Fine-looking scatterplots should not imply microscopy-level coordinates if jitter or plotting aesthetics exceed the assay’s information.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented sci-Space table

These fictional sections separate spatial and single-cell limits.

All values are invented.

Contents · Previous section · Next section

Section 16 of 36

16. Model gene counts with cell type in view

A gene may appear spatially patterned because the abundance of its cell type changes across the embryo. Test within-cell-type expression as well as composition. Otherwise, anatomy-driven cell mixtures can be mistaken for spatial regulation inside one kind of cell.

Contents · Previous section · Next section

Section 17 of 36

17. Use developmental time carefully

Pseudotime orders cells by expression similarity; it is not a stopwatch. A correlation between pseudotime and spatial position can suggest a migration or maturation pattern. Direction and mechanism require lineage, time-course or perturbation evidence beyond one developmental stage.

Contents · Previous section · Next section

Section 18 of 36

18. Quantify sampling fraction

The paper estimated that a minority of nuclei in the tissue were sampled. Publish recovered nuclei per area and the fraction passing every gate. Sparse sampling can still reveal patterns, but rare cell types or narrow boundaries may be missed unevenly.

Contents · Previous section · Next section

Section 19 of 36

19. Preserve anatomy in statistical tests

Embryo structures are not exchangeable random squares. Null models should respect sections, axes and tissue boundaries. Simple random shuffling can create misleading enrichment because neighbouring regions differ strongly in cell composition and developmental history.

Contents · Previous section · Next section

Section 20 of 36

20. Use embryos as the biological unit

Nuclei nest within sections, and sections nest within embryos. Treating every nucleus as independent inflates certainty. Summarise spatial effects by embryo and, for broad claims, include enough independently developed embryos to distinguish reproducible anatomy from one specimen.

Contents · Previous section · Next section

Section 21 of 36

21. Audit the spatial-to-sequence funnel

Count printed spots, registered spots, hashed nuclei, extracted nuclei, transcriptomes, singlets, confidently mapped cells and retained genes. This funnel shows whether a result reflects the whole tissue or a carefully selected subset of readable nuclei.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge hash diffusion

Adjacent spatial oligos can mix before or during nuclear labelling. Border controls, alternating patterns and measured spot profiles estimate crossover. When ambiguity is substantial, merge regions or label assignments probabilistically instead of insisting on a crisp boundary.

Contents · Previous section · Next section

Section 23 of 36

23. Challenge tissue deformation

Fresh-frozen sections may stretch, fold or compress between imaging and transfer. Fiducials cannot correct damage that changes local geometry unpredictably. Mark affected regions and retain the unwarped images and transforms so readers can audit anatomical assignments.

Contents · Previous section · Next section

Section 24 of 36

24. Challenge dissociation bias

Nuclei from dense, fragile or specialised tissues can be under-recovered. Compare spatially observed anatomy with the final cell-type distribution. An apparent absence after sequencing is not proof that a cell type was absent from the embryo.

Contents · Previous section · Next section

Section 25 of 36

25. Challenge reference-age mismatch

A cell atlas from adjacent developmental days helps annotation but may not contain every E14.0 state. Keep low-probability and novel profiles unresolved. Forcing each nucleus into the closest familiar label can turn developmental novelty into a false certainty.

Contents · Previous section · Next section

Section 26 of 36

26. Challenge two-embryo generalisation

Fourteen sections provide anatomical coverage, yet sections from one embryo share genetics, development and handling. Replicate conclusions across more independent embryos, litters and processing batches before treating subtle spatial differences as population-wide developmental rules.

Contents · Previous section · Next section

Section 27 of 36

27. Keep causality outside a descriptive map

Spatial co-occurrence and gene-expression gradients generate hypotheses about signals and migration. They do not prove one region instructs another. Perturbation, lineage tracing, live imaging or targeted validation is needed to test the proposed mechanism.

Contents · Previous section · Next section

Section 28 of 36

28. Compare sci-Space with ordinary scRNA-seq

Both can recover rich single-cell transcriptomes through sequencing, but sci-Space adds an approximate hash-derived region. The extra coordinate changes experimental design and uncertainty. Non-spatial atlases are useful references, yet they cannot independently confirm the new tissue position.

Contents · Previous section · Next section

Section 29 of 36

29. Run a safe paper-hash activity

Students press coloured grid stamps onto transparent paper cells, then shuffle the cells and recover both colour and written traits. Deliberate blur at grid borders shows why a spatial label can survive dissociation while remaining approximate.

Contents · Previous section · Next section

Section 30 of 36

30. Connect to Primary Science classification

Children can sort paper cells by traits and by the coloured region from which they came. They learn that two labels answer different questions: what an object is like and where it was collected.

Contents · Previous section · Next section

Section 31 of 36

31. Strengthen PSLE Science reasoning

Pupils compare the fictional rows, identify variables and explain why many recovered cells do not solve poor spatial labelling. They practise distinguishing repeat measurements from truly independent specimens and separating an observation from a proposed cause.

Contents · Previous section · Next section

Section 32 of 36

32. Extend through Secondary and O-Level Science

Biology supplies development, cells and gene expression; Chemistry supplies diffusion and molecular binding; Mathematics supplies coordinates and sampling; Computing supplies indexing and reference mapping. The method shows why modern Science questions cross subject boundaries.

Contents · Previous section · Next section

Section 33 of 36

33. Use the topic when considering schools

Check official school pages for inquiry, biology, coding, statistics and ethics opportunities. Public embryo datasets or paper models are enough for safe projects. Do not invent programmes, facilities, selection criteria or promised research access.

Contents · Previous section · Next section

Section 34 of 36

34. See the career network

Embryo-scale single-cell work connects developmental biology, genomics, microarray fabrication, imaging, computation, statistics and animal ethics. Training requirements vary by role and institution. A topic can open questions without guaranteeing a specific course or career.

Contents · Previous section · Next section

Section 35 of 36

35. Did You Know? Single-cell and precise-position are different promises

sci-Space recovers transcriptomes at the level of individual nuclei, while the spatial coordinate comes from a larger patterned region. One part can be single-cell even when the other is approximate. Careful science names both scales instead of blending them.

Contents · Previous section · Next section

Section 36 of 36

36. Close with a bounded sci-Space conclusion

A responsible statement names tissue age, animals, hash geometry, diffusion, registration, nuclear recovery, combinatorial indexing, doublets, sampling fraction, reference mapping and hierarchy. It may report single-cell transcriptomes associated with approximate embryo regions; it should not claim exact cellular coordinates, complete sampling or causal developmental mechanisms.

Pre-registration can specify embryo stage, animal count, section plane, hash-grid geometry, transfer time, fiducials, minimum hash count, transcriptome gate, collision rule, reference atlas, spatial test and independent biological unit. Exploratory cell classes and gradients can follow separately. This reduces the temptation to change labels or spatial boundaries until one developmental story appears.

The printed grid should be published as measured rather than idealised. Record spot centres, radii, missing deposits, fluorescence and overlap. Map the effective transfer footprint with alternating barcodes or inert sections. The spatial unit is the region that actually labels nuclei, which may be wider and less regular than the artwork sent to the printer.

Hash assignment needs a probabilistic audit. Release per-nucleus counts for every candidate hash, dominance ratios and ambiguity thresholds. Show where multi-hash nuclei sit relative to borders and whether they differ in transcriptome quality. Keeping ambiguous cells is scientifically useful because they reveal how the spatial labelling behaved.

The full funnel should count section area, estimated nuclei, registered spots, hashed nuclei, extracted nuclei, retained transcriptomes, singlets, confidently located cells and annotated classes. The founding paper reported an estimated sampling fraction; future studies should do the same. A spatial atlas is easier to interpret when its missing population is quantified.

Reference integration requires provenance and humility. Publish atlas version, developmental ages, preprocessing, label hierarchy and prediction scores. Keep cells unresolved when no reference class fits well. A non-spatial atlas can help name profiles, but it cannot confirm their location or guarantee that a new developmental state already exists in the dictionary.

Within-cell-type spatial expression should be separated from changing cell composition. Model cell type and region together, then report genes whose spatial association persists among comparable cells. This avoids calling a gene ‘spatially regulated’ merely because the cell population expressing it is concentrated in one organ.

Pseudotime analyses should preserve alternative explanations. A gradient may reflect migration, maturation, selective sampling or reference mismatch. Report root choice, trajectory uncertainty and sensitivity to cell filters. Live imaging, lineage tracing or perturbation is needed when the biological claim uses directional words such as moves, becomes or instructs.

Statistics must reflect embryos and sections. Adjacent sections share one organism, while cells share preparation and developmental history. Use multilevel models or specimen summaries and show embryo-specific effects. Thousands of cells can make a tiny, embryo-specific difference look numerically certain if the hierarchy is ignored.

Protocol transfer needs new optimisation for organ, stage and preservation. Later embryos, adult tissues and diseased specimens differ in thickness, matrix and nuclear release. Validate hash diffusion, recovery and reference fit each time. The method’s logic transfers more readily than its numerical performance.

A careful conclusion might read: ‘Across independently processed embryo sections, sci-Space recovered single-nucleus transcriptomes associated with registered hashing regions after transfer, ambiguity, dissociation, collision, sampling and reference-mapping controls.’ It should name embryos separately. It should not claim exact cellular coordinates or complete representation of all nuclei.

Reproducibility requires printing files, oligo sequences, fiducial images, grid measurements, transfer conditions, raw hashes, nuclear transcriptomes, doublet scores, masks, transforms, atlas objects, label probabilities, code, environments and checksums. Failed sections and unmatched cells should remain available. The next experiment should reduce the largest diffusion, recovery or generalisation uncertainty.

Figures should pair embryo anatomy with measured grid spots, border ambiguity, cell-recovery maps, annotation confidence and embryo-level summaries. A developmental trajectory figure should show spatial bins and uncertainty rather than only a smooth arrow. The visual design must not promise more coordinate precision or temporal direction than the assay observed.

Section orientation needs a durable record. Sagittal sections from related positions can still contain different structures and cell mixtures. Store block-face images, section order and anatomical annotations, then compare like with like across embryos. A pooled atlas may look continuous even when coverage differs systematically between specimens.

Sensitivity analyses should repeat central findings after excluding border hashes, tightening doublet filters and changing the reference mapping threshold. A robust spatial programme should remain recognisable. A result that depends on ambiguous borders or one atlas setting belongs in the exploratory column.

Gene-expression depth should be reported with dropout rather than one median alone. Plot genes and molecules by cell class, section and hash confidence. Low-depth nuclei can still carry useful broad identities, but fine subtype or pathway claims need enough information to distinguish absence from non-detection.

Contents · Previous section · Continue to the Science Learning Hub