Why Science? | sci-CAR, Combinatorial Indexing and Chromatin-Accessibility–RNA Evidence

Section 1 of 36

1. Ask a two-layer response question

sci-CAR is useful when the question needs chromatin accessibility and RNA from the same cell. In a treatment response, one can ask which sites change accessibility, which genes change transcript abundance and whether both changes occur in the same cellular state. That is more precise than calling every nearby open region a regulator.

Contents · Next section

Section 2 of 36

2. Did you know an index can become a cell address?

A nucleus visits more than one barcoding reaction. The ordered combination of indexes becomes an address shared by its accessible-DNA and RNA-derived molecules. Combinatorial indexing gains throughput without placing every starting cell in a separate tube, but it depends on accurate wells, balanced loading and sufficiently distinct barcode sequences.

Contents · Previous section · Next section

Section 3 of 36

3. Define accessible chromatin conservatively

Tn5 transposase inserts adaptors where DNA is reachable under the assay conditions. Enrichment near transcription start sites can indicate useful signal, yet accessibility does not prove enhancer activity, transcription-factor binding or causal regulation. Nucleosome organisation, sequence preference, permeabilisation and enzyme concentration all influence the fragments that appear.

Contents · Previous section · Next section

Section 4 of 36

4. Treat RNA counts as samples

Reverse transcription captures only a portion of the RNA available in each nucleus or cell. Low-abundance transcripts may be missed, while abundant or stable transcripts are easier to detect. A zero therefore means not detected, not necessarily absent. Molecule counts, detected genes and ambient-RNA estimates belong beside any biological interpretation.

Contents · Previous section · Next section

Section 5 of 36

5. Understand the same-cell advantage

When both libraries inherit one cellular address, researchers can compare modalities without matching separate ATAC-only and RNA-only cells after the experiment. This reduces one source of uncertainty. It does not remove the need to prove that the address represents one good cell and that both channels are informative.

Contents · Previous section · Next section

Section 6 of 36

6. Keep correlation below causation

Accessibility and expression may covary because a distal element regulates a gene, because both respond to treatment, or because cell composition changes. Genomic distance and correlation suggest candidates. Temporal sampling, perturbation, reporter assays or orthogonal measurements are needed to distinguish direct regulation from shared context.

Contents · Previous section · Next section

Section 7 of 36

7. Design biological replication first

List independent cultures or animals, treatment doses, time points, expected states and primary comparisons before counting cells. Spread conditions across processing batches. Thousands of cells from one preparation can map heterogeneity, but they do not substitute for independent specimens when the conclusion concerns treatment, tissue or population-level effects.

Contents · Previous section · Next section

Section 8 of 36

8. Start the RNA index in situ

A poly(T) reverse-transcription primer carrying a well-specific index and unique molecular identifier copies polyadenylated RNA inside permeabilised material. The UMI helps recognise amplification copies. RNA integrity, priming efficiency, incubation and leakage all affect the channel, so spike-ins and plate-position summaries can reveal uneven chemistry.

Contents · Previous section · Next section

Section 9 of 36

9. Add the accessibility index

A barcoded Tn5 complex cuts and tags accessible DNA. The reaction must expose chromatin while preserving indexed cDNA and nuclear identity. Report enzyme lot, concentration, time, temperature and input. A plate effect at this step can later appear as a biological accessibility programme if well positions are ignored.

Contents · Previous section · Next section

Section 10 of 36

10. Pool, redistribute and finish the address

After the first molecular indexes, nuclei are pooled and sorted into new wells by FACS. Later amplification indexes expand the address. This design creates many combinations from a manageable number of reactions. Plate maps and sorting records are therefore primary evidence, not merely laboratory administration.

Contents · Previous section · Next section

Section 11 of 36

11. Split lysate into modality libraries

The indexed material is lysed and divided so RNA-derived cDNA and ATAC-derived DNA can be amplified separately. Separation allows each library to use appropriate primers and analysis. It can also create unequal recovery. The accepted joint-cell set should require declared minimum evidence in both channels, not a convenient union.

Contents · Previous section · Next section

Section 12 of 36

12. Plan sequencing with a pilot

Estimate unique ATAC fragments, RNA molecules, genes, duplicate rates and saturation before scaling. More reads help only while libraries contain unseen molecules. More cells help resolve diversity; more biological replicates test reproducibility. Those three investments answer different questions and should not be combined into one headline number.

Contents · Previous section · Next section

Section 13 of 36

13. Demultiplex without inventing identity

Declare exact-match and mismatch policies for every index, retain index quality and measure reads assigned to unused combinations. Permissive rescue can move molecules from an abundant cell into a weak barcode. Repeat major results under strict matching and publish how many cells or molecules depend on rescue.

Contents · Previous section · Next section

Section 14 of 36

14. Detect collisions and doublets separately

A barcode collision assigns two nuclei the same combinatorial address; a doublet carries two nuclei together. Both may create hybrid cell states. Barcode-occupancy models, species-mixing experiments, unusually high counts and incompatible lineage markers detect different aspects. Report estimated rates and repeat headline analyses after excluding high-risk profiles.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with fictional paired profiles

These invented values are for classroom comparison only, not performance benchmarks.

Contents · Previous section · Next section

Section 16 of 36

16. Audit two quality funnels

For ATAC, show reads, mapping, duplicates, accepted fragments, transcription-start-site enrichment and nucleosomal pattern. For RNA, show mapping, UMIs, genes and ambient estimates. Then show the intersection. A single composite score may hide one excellent channel paired with one failed channel, exactly where a joint claim becomes unsafe.

Contents · Previous section · Next section

Section 17 of 36

17. Use the dexamethasone series carefully

The foundational sci-CAR study analysed 4,825 cells across a dexamethasone response series. Time and treatment can help organise response states, but each time point needs replication and balanced handling. A trajectory through pooled cells is not equivalent to watching one cell change, and response-associated links remain candidates.

Contents · Previous section · Next section

Section 18 of 36

18. Learn from the adult kidney atlas

The same study profiled 11,296 adult mouse kidney cells, demonstrating that paired measurements can distinguish tissue populations and relate regulatory accessibility to transcription. Cell abundance, dissociation sensitivity and depth differ among kidney cell types. Rare clusters should recur across animals and survive conservative doublet filters.

Contents · Previous section · Next section

Section 19 of 36

19. Cluster each modality before integration

Build RNA-only and ATAC-only views before a joint embedding. Overlay specimen, plate, depth and cell-cycle information. Disagreement may reveal a biological transition, but it may also show channel failure or mixture. Integration should explain discordance rather than force every profile into one attractive map.

Contents · Previous section · Next section

Section 20 of 36

20. Link sites and genes with explicit nulls

Candidate cis-regulatory links can be based on genomic distance and accessibility–expression covariance. Match background sites for distance, variability and accessibility; control depth and state composition; adjust multiple tests. Show raw distributions and independent validation. A nearby correlation is a hypothesis generator, not a certified enhancer–gene connection.

Contents · Previous section · Next section

Section 21 of 36

21. Estimate treatment effects at the specimen level

Cells are observations nested inside biological replicates. Aggregate or model them accordingly when testing treatment differences. Balance cell counts, report effect sizes and uncertainty, and repeat after downsampling dominant states. A tiny cell-level p-value can coexist with poor replicate agreement, which is the result readers need to see.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge sparse zeros

Missing fragments and transcripts often arise from sampling. Aggregate only across biologically coherent units, report cells and specimens behind every curve, and downsample high-depth groups. A relationship that appears only after aggressive smoothing should be labelled exploratory until a targeted experiment measures it more directly.

Contents · Previous section · Next section

Section 23 of 36

23. Challenge cell-state composition

Treatment may change the proportion of states rather than the molecular programme within a state. Report composition separately from within-state responses. Reweight or stratify profiles, retain replicate structure and explain exclusions. Otherwise an accessibility–RNA association may simply compare different populations.

Contents · Previous section · Next section

Section 24 of 36

24. Challenge transposase and sequence bias

Tn5 insertion preference, mappability and repetitive sequence shape accessible-DNA recovery. Compare candidate sites with matched background, inspect raw fragments and use orthogonal accessibility or occupancy evidence where decisive. The chemistry is wonderfully useful, but it is not a neutral camera pointed at the genome.

Contents · Previous section · Next section

Section 25 of 36

25. Challenge time-order stories

If accessibility appears before RNA, state the sampling times, trajectory root, smoothing and uncertainty. Repeat across replicates and alternative models. Temporal precedence strengthens a regulatory hypothesis, yet both layers may respond to an unmeasured driver. A perturbation is required to test direction.

Contents · Previous section · Next section

Section 26 of 36

26. Validate the weakest inference

Choose follow-up according to the claim: targeted accessibility for a signal, independent RNA measurement for expression, chromatin-contact data for physical proximity, or CRISPR and reporter experiments for function. Repeating only the strongest part of the evidence leaves the decisive causal gap untouched.

Contents · Previous section · Next section

Section 27 of 36

27. Report the evidence ladder

Separate valid cellular address, adequate ATAC, adequate RNA, reproducible state, robust association, temporal ordering and functional validation. Each rung supports the next but does not guarantee it. Readers should be able to see where the published result stops and which experiment would move it upward.

Contents · Previous section · Next section

Section 28 of 36

28. Build Primary Science observation habits

Children can sort cards into what was observed and what is an explanation: open region, RNA count, cell label and proposed cause. This turns an advanced assay into a familiar discipline—describe first, infer second. The method name matters less than learning to keep evidence and story in separate boxes.

Contents · Previous section · Next section

Section 29 of 36

29. Prepare for PSLE Science data questions

Plot two fictional measurements for the same set of cells and ask which change is larger, which comparison is fair and what could produce an outlier. Students practise reading tables, controlling variables and recognising that a pattern can support more than one explanation.

Contents · Previous section · Next section

Section 30 of 36

30. Deepen Secondary Science reasoning

Map the workflow as specimen, enzyme, barcode, sequence, table and claim. At each arrow, name a possible loss or mix-up and one control. This connects enzymes, nucleic acids, variation and reliability to a real research system without requiring learners to memorise computational jargon.

Contents · Previous section · Next section

Section 31 of 36

31. Connect to O-Level Science

The official 2026 Singapore–Cambridge Biology and Chemistry syllabuses emphasise evidence-based explanation, planning, analysis and evaluation. sci-CAR provides a modern context for enzyme specificity, chemical conditions, biological organisation and measurement limits. Use it to practise reasoning, not to collect fashionable terms.

Contents · Previous section · Next section

Section 32 of 36

32. Ask school-choice questions about fit

Families can ask how a school develops practical inquiry, data literacy, computing, communication and mentoring. Verify current programmes on official school pages and at open houses. No single research technique proves a school is suitable; sustained learning culture, access and the student’s interests matter.

Contents · Previous section · Next section

Section 33 of 36

33. See a broad career ecosystem

Joint-omics work involves molecular biologists, laboratory technologists, computational scientists, statisticians, software engineers, instrument specialists, data stewards and science communicators. Students may enter through junior college, polytechnic and other routes. Careers grow through transferable skills and further training, not a guaranteed linear outcome.

Contents · Previous section · Next section

Section 34 of 36

34. Know when science tuition helps

Science tuition can help when it diagnoses a specific gap such as graph reading, experimental design, precise vocabulary or evaluation of evidence. It should not replace school feedback, sleep or independent practice. Useful lessons ask learners to transfer the reasoning to a new dataset rather than recite a polished answer.

Contents · Previous section · Next section

Section 35 of 36

35. Did you know the method connected two atlases?

sci-CAR demonstrated the same indexing logic in a controlled dexamethasone series and in adult mouse kidney tissue. That range is exciting because it shows how one measurement framework can address response and cell identity. It also reminds us that every new tissue needs fresh quality checks and replication.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a robustness grid

Vary barcode correction, collision filters, ATAC and RNA thresholds, feature sets, integration methods, replicate inclusion and site–gene rules. Mark which states and links persist. Stable findings deserve stronger language; setting-dependent findings remain worthwhile hypotheses with an honest next test.

A publication-ready sci-CAR study should preserve the complete address ledger. For every cellular index combination, retain the contributing wells, index qualities, sorting destination, molecule counts and reason for acceptance or exclusion. Publish unused and low-count combinations as part of the background model. The same information that makes the method scalable also makes its failure modes diagnosable. Without the address ledger, a reader cannot distinguish a rare state from a rescued index or plate-local contamination.

Treatment experiments need a sample sheet that separates dose, elapsed time, preparation batch and biological replicate. Randomise conditions across plates and avoid processing all controls before all treated material. Model cells within replicates rather than treating every cell as independent. Show condition effects in each replicate and include effect sizes. If a response exists only in one preparation, it may be a useful lead, but it is not yet a general treatment programme.

Kidney applications require tissue-aware recovery checks. Compare cell-type proportions with histology, established markers or an independent reference, and explain which populations may be fragile during dissociation. Report animals, regions and cells contributing to each cluster and regulatory link. A cell type represented by many cells from one animal is not broadly replicated. Rare populations should persist under conservative collision removal and after leaving out each animal.

Site–gene links deserve a transparent candidate table. Give genomic distance, effect direction, accessibility coverage, RNA coverage, cell states, replicate agreement, multiple-testing result and any contact or perturbation evidence. Separate promoter-proximal and distal rules. Readers should be able to identify whether a link is driven by abundant states, one time point or uneven depth. This converts an attractive network into an auditable set of hypotheses.

Response timing should use observed time as a check on inferred order. Estimate change points separately for accessibility and RNA, carry uncertainty forward and repeat under alternative smoothing. Compare the same candidate across independent cultures. An early accessibility shift followed by RNA is evidence consistent with priming, while simultaneous response or replicate disagreement suggests other models. Phrase each result at the level supported by the analysis.

Figures should move from specimen balance and index quality to two-channel performance, independent cell-state views, replicate-aware response effects and candidate regulatory links. Show raw fragments and RNA molecules beneath smooth trends. Display negative examples and ambiguous cells as well as the clean story. The layout becomes more persuasive because readers can see exactly how every inference was earned.

Reproducibility includes fixation or permeabilisation, reverse-transcription and Tn5 conditions, oligonucleotide sequences, FACS settings, plate maps, read structures, barcode correction, molecule counting, cell calls, peak sets, joint matrices, models, code and environments. Genomic data need proportionate governance. Safety covers biological material, fixatives, detergents, enzymes, lasers, sharps, heat and amplified DNA under institutional procedures.

Close the investigation with a decision table. For each claim, list the observation, competing explanations, sensitivity checks, independent replicate evidence and next discriminating experiment. A result that survives strict barcodes, balanced depth and leave-one-replicate-out analysis can be described confidently. A result that changes with one setting should remain visible as an exploratory hypothesis. That is not a failure; it is science identifying where new evidence is most valuable.

Contents · Previous section · Continue to the Science Learning Hub