Why Science? | MERFISH, Error-Robust Barcodes and Spatial RNA Evidence

Section 1 of 36

1. Begin with RNA molecules that keep their place

MERFISH detects selected RNA species inside fixed cells or tissue. Encoding probes bind target transcripts, and sequential fluorescent readouts identify a designed barcode. Each decoded spot retains image coordinates, connecting molecular identity with location.

Contents · Next section

Section 2 of 36

2. Did you know one gene can be a binary word?

A target receives a pattern of on and off readout bits across imaging rounds. The codebook separates valid words so that some missing or false bits can be corrected. Molecular biology and information theory meet in one microscope.

Contents · Previous section · Next section

Section 3 of 36

3. Use error-robust codes deliberately

Code distance determines how many bit errors can be detected or corrected. More targets require enough barcode space and balanced bit usage. The codebook is part of the assay, not a hidden software detail.

Contents · Previous section · Next section

Section 4 of 36

4. Distinguish encoding probes from readout probes

Encoding probes bind RNA and carry readout sequences. Fluorescent readout probes bind those sequences during each cycle. Separating target recognition from colour readout enables multiplexing but creates two hybridisation steps to validate.

Contents · Previous section · Next section

Section 5 of 36

5. Treat the gene panel as a sampling frame

MERFISH often targets a chosen panel rather than every transcript. Panel design should match the biological question, include controls and distinguish expected cell states. Unmeasured genes cannot be recovered from image beauty.

Contents · Previous section · Next section

Section 6 of 36

6. Understand optical crowding

When many molecules occupy a small region, fluorescent spots overlap and decoding becomes difficult. Tissue clearing, expansion, z-stacks or sparser panels can help. Resolution and molecule density jointly define usable multiplexing.

Contents · Previous section · Next section

Section 7 of 36

7. Define the spatial question first

Specify whether the endpoint is molecule localisation, cell types, tissue domains or neighbourhoods. Predefine specimens, regions, panel and validation. A rainbow map should not choose the hypothesis after imaging.

Contents · Previous section · Next section

Section 8 of 36

8. Preserve morphology and RNA

Fixation, freezing, section thickness and handling affect both tissue shape and transcript retention. Optimise on representative material and record delays. A crisp image with lost RNA, or high counts with distorted anatomy, is incomplete evidence.

Contents · Previous section · Next section

Section 9 of 36

9. Design probes for specificity

Avoid repeats, close paralogues and inaccessible regions where possible. Use multiple encoding probes per transcript and test off-target binding. Probe number and sequence composition affect brightness and detection efficiency.

Contents · Previous section · Next section

Section 10 of 36

10. Control autofluorescence

Tissues contain natural fluorophores whose signal can vary by wavelength and treatment. Measure no-probe or blank-code controls, use compatible clearing and choose channels carefully. Background that changes across anatomy can mimic spatial expression.

Contents · Previous section · Next section

Section 11 of 36

11. Anchor RNA and clear tissue carefully

Crosslinking or gel embedding can retain molecules while proteins and lipids are removed to improve imaging. Over-clearing damages morphology; under-clearing raises scatter and background. Record expansion or deformation when it occurs.

Contents · Previous section · Next section

Section 12 of 36

12. Choose fiducials for registration

Fluorescent beads or stable cellular features help align repeated imaging rounds. Fiducials should cover the field and remain visible. Global alignment can still fail locally when tissue warps.

Contents · Previous section · Next section

Section 13 of 36

13. Balance imaging rounds

Exposure, laser power, focus and readout concentration should be consistent. Bleaching and optical drift accumulate across cycles. Randomise bit assignments so late rounds do not systematically disadvantage one biological class.

Contents · Previous section · Next section

Section 14 of 36

14. Protect against carryover

Readout probes must be removed or fluorescence extinguished before the next round. Incomplete stripping creates false positive bits; harsh stripping damages the sample. Measure residual signal after representative cycles.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented MERFISH table

These fictional bit calls teach decoding, not diagnosis.

Contents · Previous section · Next section

Section 16 of 36

16. Register rounds before decoding

Estimate translation, rotation and local deformation from fiducials. Inspect residual displacement relative to spot size. A one-pixel error can change barcode identity when molecules are dense.

Contents · Previous section · Next section

Section 17 of 36

17. Detect spots with a declared threshold

Spot-finding algorithms trade missed dim molecules against false background calls. Thresholds should use controls and remain consistent across groups. Report sensitivity rather than tuning each image to look equally populated.

Contents · Previous section · Next section

Section 18 of 36

18. Decode against the codebook

Match observed bit patterns to valid codes under a stated error rule. Publish counts of exact, corrected and rejected words. Error correction is powerful only when the empirical error model supports it.

Contents · Previous section · Next section

Section 19 of 36

19. Segment cells with uncertainty

Nuclear and membrane stains help assign molecules to cells, but boundaries overlap and sections truncate cells. Show overlays and quantify unassigned molecules. Cell-level expression inherits segmentation error.

Contents · Previous section · Next section

Section 20 of 36

20. Normalise without erasing anatomy

Cell size, section depth and detection efficiency change counts. Use appropriate offsets or controls and examine raw molecule maps. Spatial gradients should not be created by local image quality.

Contents · Previous section · Next section

Section 21 of 36

21. Replicate across specimens

Millions of spots within one section remain one biological specimen. Show that key domains or cell associations recur across independent tissues. Spatial correlation must be included in inference.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge whole-transcriptome language

A large targeted panel can describe many states, but genes outside the codebook remain unmeasured. Call the assay targeted when that is what it is, and use sequencing for broader discovery when needed.

Contents · Previous section · Next section

Section 23 of 36

23. Challenge perfect error correction

Codes can correct a limited number of independent bit errors. Systematic focus loss, spectral bleed or registration drift may violate the code model. Publish bit-wise error rates and blank-code behaviour.

Contents · Previous section · Next section

Section 24 of 36

24. Challenge molecule-count certainty

Hybridisation efficiency differs among targets, and overlapping spots may be missed. Counts are informative but not automatically equal-efficiency absolute molecules across genes. Validate representative targets with single-molecule FISH.

Contents · Previous section · Next section

Section 25 of 36

25. Challenge cell-neighbour claims

Two segmented cells may touch in a two-dimensional section without being neighbours in the intact volume. Section thickness, boundary error and three-dimensional anatomy limit causal interpretation.

Contents · Previous section · Next section

Section 26 of 36

26. Challenge panel-driven cell types

A clustering model can only distinguish states represented by the selected genes. Missing markers may merge or mislabel populations. Compare with an independent single-cell reference and retain uncertainty.

Contents · Previous section · Next section

Section 27 of 36

27. Validate with orthogonal images

Use conventional smFISH, immunofluorescence, histology or another spatial platform for central targets and boundaries. Agreement should be assessed with coordinates and counts, not only visual similarity.

Contents · Previous section · Next section

Section 28 of 36

28. Learn MERFISH evidence with a safe classroom model

Students can give fictional RNA cards binary words, flip one bit and test whether the codebook still identifies the target. A paper grid then preserves location, teaching error correction and spatial provenance without probes or microscopes.

Contents · Previous section · Next section

Section 29 of 36

29. Build Primary Science process skills

Young learners can identify the sample, label, measured signal and comparison. They can explain why repeated samples, a known reference and a clearly recorded procedure make an observation more trustworthy.

Contents · Previous section · Next section

Section 30 of 36

30. Prepare for PSLE Science reasoning

A simplified MERFISH evidence dataset can practise fair comparisons, pattern reading and evidence-linked explanations. It is enrichment rather than examined content. The reusable skill is to separate the direct observation from the inference built on it.

Contents · Previous section · Next section

Section 31 of 36

31. Extend into Secondary and O-Level Science

Biology, Chemistry, Mathematics and Computing meet in MERFISH evidence. Students can connect cells and molecules to reactions, graphs, probability and data handling while keeping variables and limitations visible.

Contents · Previous section · Next section

Section 32 of 36

32. Use the topic for school choices

Families can ask how a school develops laboratory safety, biological reasoning, data literacy and communication. Verify current official programmes and entry requirements. Strong foundations matter more than access to one specialist instrument.

Contents · Previous section · Next section

Section 33 of 36

33. See the career ecosystem without promises

MERFISH connects molecular biology, microscopy, optics, information theory, image analysis, developmental biology and pathology research. Roles and qualifications vary; credible maps require both careful chemistry and transparent computation.

Contents · Previous section · Next section

Section 34 of 36

34. Use questions for science tuition and enrichment

Ask why one failed imaging round does not always ruin a barcode, how registration error changes identity, and why a targeted panel cannot support claims about every gene.

Contents · Previous section · Next section

Section 35 of 36

35. Did you know this measurement hides a second story?

A blank barcode is useful. Blank codes and unassigned patterns reveal empirical false-decoding rates, turning deliberately absent targets into controls for the entire imaging-and-decoding chain.

Contents · Previous section · Next section

Section 36 of 36

36. Conclude with a MERFISH evidence checklist

Before accepting a MERFISH map, ask: Were probes specific and tissue preserved? Were cycles registered, carryover controlled and bit errors measured? Were cells segmented, specimens replicated and key targets validated? Then optical barcodes can support spatial RNA evidence.

A defensible MERFISH project starts with a gene-panel and geometry plan. Define specimen, region, section thickness, optical volume, target genes, blank codes, expected molecule density, barcode distance, imaging rounds and smallest spatial pattern worth detecting. Simulate code capacity and spot density before ordering probes. A panel should include biological markers, abundance ranges, housekeeping references and controls that can reveal false decoding.

Encoding-probe design should record transcript reference, targeted regions, specificity filters, melting properties, readout sequences and probe count per gene. Avoid repeats and close paralogues where possible. Test representative high-, medium- and low-abundance targets with conventional smFISH. A missing signal may reflect inaccessible target sequence rather than absent RNA, while off-target binding can create spatially convincing molecules.

Tissue preparation must preserve both RNA and anatomy. Record collection delay, fixation or freezing, embedding, section thickness, storage and clearing. Optimise on adjacent representative sections, not the only experimental specimen. Inspect morphology and RNA retention together. If hydrogel expansion or aggressive clearing changes dimensions, measure deformation and carry the transform into every spatial claim.

Imaging cycles need a run sheet with readout probe, fluorophore, exposure, laser power, focus, stripping condition and elapsed time. Randomise code bits across early and late rounds so photobleaching or tissue damage does not target one gene class. Use fiducials distributed across each field and inspect local registration residuals. A small average error can hide a failed corner of the image.

Spot detection should be calibrated with no-probe images, blank codes and known targets. Declare threshold, point-spread assumptions and z-stack handling. Dense regions may merge spots; dim regions may fall below detection. Show precision-recall or sensitivity analyses when benchmark data exist. Do not tune thresholds separately for groups to produce visually balanced maps.

Decoding should publish exact, corrected and rejected code counts, bit-wise dropout, false-positive bits and blank-code rate. Error-correcting codes work under a model; systematic failure of one imaging round can affect many genes together. Compare decoded abundance with bulk RNA or sequencing references while remembering that different platforms have different capture efficiencies.

Cell segmentation should preserve uncertainty. Use nuclei, membranes or morphology with a declared algorithm and parameters. Show representative overlays across crowded, sparse and damaged regions. Report molecules outside cells and sensitivity to boundary changes. A cell-type label inherits both the gene-panel limitation and the segmentation error, so confidence should not be expressed more precisely than the inputs.

Spatial analysis must treat neighbouring molecules and cells as correlated. Biological specimens remain the independent units. Test domains and neighbourhoods across sections and subjects, use physical scale bars and avoid interpreting thousands of cells from one section as thousands of tissues. When a pattern depends on smoothing width or neighbour radius, show that dependence explicitly.

Figures should pair raw images, fiducials, registration residuals, blank-code controls, spot examples, bit-error summaries, molecule maps, segmentation overlays, specimen-level replication and orthogonal validation. Use common colour limits. A polished map without raw spot evidence makes it difficult to distinguish biology from decoding and display choices.

A useful claim audit begins with one decoded molecule and works outward: observed pixels, registered spot, bit pattern, matched code, segmented cell, cell label, spatial domain and biological interpretation. Record the uncertainty introduced at each step. Re-run the central field from raw images in a clean environment with pinned codebook and thresholds, then perturb registration, spot detection and cell boundaries within defensible ranges. If the tissue pattern vanishes after a one-pixel shift or a small threshold change, it is not yet a durable finding. This audit also gives learners a memorable lesson: error correction can repair limited noise, but it cannot rescue a measurement model whose assumptions were never tested.

Panel design should be judged against the intended spatial claim. A compact set of lineage markers may separate major cell classes but cannot automatically resolve transient states or mechanisms. Conversely, adding many genes can increase optical crowding and reduce reliable decoding in dense regions. Report why each gene is present, which alternative explanations it helps test and which biology remains invisible. This turns the codebook into an explicit scientific argument instead of an impressive list of targets, and it keeps later cell labels within the actual resolving power of the experiment.

Replication must preserve anatomical context without confusing repeated images for repeated biology. Multiple fields from one section can reveal local variation, and serial sections can test registration or continuity, but the independent unit is still the specimen that generated them. Predefine how regions are sampled, document missing or damaged fields and summarise effects at specimen level. When a striking neighbourhood appears in one field, search for it in held-out sections and subjects before naming it as a general tissue programme. Spatial richness increases the need for careful replication; it does not replace it.

Display choices deserve the same discipline. Colour scales, point size, smoothing and transparency can make sparse molecules look like a continuous domain or hide a dense boundary. Pair overview maps with unsmoothed local views and counts per physical area, using identical scales across conditions. Include coordinates and scale bars so that a reader can relate the pattern to tissue structure rather than to the canvas. A spatial conclusion should remain recognisable when the palette changes and when the underlying molecules, not only the rendered surface, are shown.

Data stewardship should preserve consent, specimen orientation, preparation logs, probe sequences, codebook, imaging order, raw image stacks, flat-field and chromatic corrections, fiducials, transforms, spot calls, decoded molecules, masks, segmentations, analysis code and final tables. Human tissue images and molecular maps can be sensitive and need explicit access and retention plans.

Safety includes tissues, fixatives, acrylamide or clearing reagents, lasers, microscopy systems, glass and genomic information. Trained staff should follow institutional procedures. Classroom learning should use paper barcodes and synthetic coordinates. The next experiment should target the dominant uncertainty: redesign weak probes, reduce optical crowding, improve registration, validate segmentation or repeat a spatial association across independent specimens. MERFISH is strongest when error correction supports—not replaces—transparent optical and biological controls.

Contents · Previous section · Continue to the Science Learning Hub