Why Science? | sci-Hi-C, Combinatorial Indexing and High-Throughput Single-Cell Genome-Contact Evidence

Section 1 of 36

1. Start with the scaling problem

Conventional single-cell Hi-C can require physical isolation and separate handling for each nucleus. sci-Hi-C asks whether successive barcode combinations can identify many cells inside a shared workflow. The answer depends on molecular indexing, not on assuming that every barcode combination belongs to exactly one clean cell.

Contents · Next section

Section 2 of 36

2. Did you know two ordinary index rounds create many identities?

If nuclei are divided across one set of wells, tagged, pooled and divided again, the ordered pair of well identities creates a much larger address space. This combinatorial idea enabled the foundational sci-Hi-C study to report 10,696 single-cell contact maps without 10,696 separate early reactions.

Contents · Previous section · Next section

Section 3 of 36

3. Separate throughput from depth

Processing more cells improves the view of population diversity, but it does not guarantee many contacts in each cell. A study may be excellent for classifying cell states yet too sparse for a detailed three-dimensional reconstruction of one nucleus. Cell count and unique contacts per cell answer different questions.

Contents · Previous section · Next section

Section 4 of 36

4. Keep Hi-C chemistry underneath the indexes

Fixed nuclei are digested, DNA ends are processed, nearby fragments are ligated and the resulting chimeras are sequenced. Barcodes tell the pipeline which cell likely contributed a read pair. They do not remove digestion, ligation, amplification, mapping or distance-dependent contact biases.

Contents · Previous section · Next section

Section 5 of 36

5. Define a barcode collision

A collision occurs when material from more than one nucleus receives the same usable barcode combination or when index errors merge identities. The resulting profile may look unusually rich or biologically intermediate. Species-mixing controls, per-cell contact counts and incompatible chromosome patterns help estimate this failure.

Contents · Previous section · Next section

Section 6 of 36

6. Define a doublet separately

A doublet is two nuclei that travel together as one physical event, whereas a barcode collision can join separately processed nuclei computationally. Both mix cells, but their causes and remedies differ. Sorting, nuclear imaging, species mixing and barcode-design calculations should therefore be reported together.

Contents · Previous section · Next section

Section 7 of 36

7. Plan the address space before the experiment

Choose well counts, expected nuclei per round and index sequences so the number of possible combinations comfortably exceeds the recovered cells. Balance index usage and edit distance. A theoretical barcode count is not the same as an effective count once failed wells, sequencing errors and uneven occupancy are included.

Contents · Previous section · Next section

Section 8 of 36

8. Fix and digest intact nuclei consistently

Crosslinking preserves proximal chromatin, while restriction digestion opens defined DNA ends. Over-fixation can reduce cutting; under-fixation can lose contacts. Record concentration, time, quench, enzyme lot, digestion efficiency and nuclear integrity for every biological preparation rather than relying on pooled averages.

Contents · Previous section · Next section

Section 9 of 36

9. Apply the first indexed bridge adaptor

After digestion and end processing, nuclei are distributed so each well receives a distinct indexed adaptor. The adaptor participates in marking proximity products. Free adaptor, unequal ligation and well contamination can skew cell recovery, so index balance and invalid junctions belong in early quality control.

Contents · Previous section · Next section

Section 10 of 36

10. Pool without losing the nuclei

The first-round wells are combined while nuclei remain physically intact. Broken nuclei release DNA that can ligate or redistribute independently of the original cell. Microscopy, supernatant DNA and contact-distance patterns help show whether pooling preserved the object that the cellular barcode is supposed to represent.

Contents · Previous section · Next section

Section 11 of 36

11. Perform proximity ligation

DNA fragments held near one another by crosslinked nuclear organisation are joined. Ligation frequency is influenced by distance, accessibility and chemistry, not solely by a stable biological loop. Controls should distinguish successful cis-rich nuclear ligation from random solution contacts and incomplete ends.

Contents · Previous section · Next section

Section 12 of 36

12. Redistribute nuclei for the second index

Pooled nuclei are split into another plate or well set. A second index added during later library construction combines with the first to identify the cell. Accurate plate maps and unique index sequences are essential; one swapped column can relabel thousands of inferred cellular profiles.

Contents · Previous section · Next section

Section 13 of 36

13. Reverse crosslinks and amplify cautiously

After cellular identities are encoded, DNA is purified, converted into sequencing libraries and amplified. Too many cycles enrich duplicates and can exaggerate a small set of molecules. Report unique complexity, saturation and contact yield per barcode, not just the total number of reads generated.

Contents · Previous section · Next section

Section 14 of 36

14. Sequence enough to evaluate the intended question

A pilot can estimate usable barcode combinations, unique contacts and collision rate. Deepening a saturated library may add duplicates rather than information. Decide whether the aim is broad cell-state discovery, chromosome-level structure or rare-state detection, then allocate reads and biological replicates accordingly.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented sci-Hi-C table

These fictional values teach quality review; they are not performance benchmarks.

Contents · Previous section · Next section

Section 16 of 36

16. Demultiplex in the declared order

Read structures should specify where each index sits, how many mismatches are allowed and how invalid combinations are treated. Preserve raw index qualities and an undetermined bin. Aggressive rescue of damaged barcodes can increase cell count while silently merging cells or moving molecules between them.

Contents · Previous section · Next section

Section 17 of 36

17. Map contacts with a transparent funnel

Trim adaptors, align both genomic ends, apply mapping-quality rules and classify dangling ends, self-circles, duplicates and valid pairs. Use a sample-aware reference when karyotypes or structural variants matter. Publish counts from raw reads to the final per-cell contact list.

Contents · Previous section · Next section

Section 18 of 36

18. Attach quality metrics to every cell

Useful fields include reads, valid contacts, duplicates, cis fraction, distance distribution, chromosome coverage, index combination, processing batch and species assignment where available. Plot distributions before choosing thresholds. A single pooled pass rate can hide one failed plate row or biological group.

Contents · Previous section · Next section

Section 19 of 36

19. Use contact sparsity honestly

Failure to observe a contact in one cell is weak evidence of absence. Domain-scale summaries, chromosome territories or cell-cycle trends can emerge across many sparse profiles. If contacts are imputed or smoothed, show raw results too and state which conclusions depend on the computational completion.

Contents · Previous section · Next section

Section 20 of 36

20. Infer cell states without circular proof

Contact features can organise cells along a cell-cycle trajectory, but the same features should not define and then validate the trajectory. Independent DNA-content measurements, known replication patterns or held-out contact summaries provide stronger checks. Technical gradients across wells must also be excluded.

Contents · Previous section · Next section

Section 21 of 36

21. Aggregate only cells that belong together

Pooling contacts after classification can create useful group maps. It can also manufacture a clean average from heterogeneous or misclassified cells. Report the classifier, uncertainty, minimum contacts, specimen composition and sensitivity to alternative clusters before treating the aggregate as a cell-type structure.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge the promised scale

A very large cell total can be driven by shallow profiles, one preparation or relaxed filters. Report unique cells after collision and doublet removal, median and distribution of contacts, independent specimens and index occupancy. Scale is scientific value only when the recovered units are interpretable.

Contents · Previous section · Next section

Section 23 of 36

23. Challenge barcode cross-talk

Index hopping, synthesis errors and carry-over can move reads into plausible combinations. Use unused barcode combinations, edit-distance-aware designs and negative wells to estimate cross-talk. Compare conclusions under stricter exact-match demultiplexing so rescued reads do not create a rare population.

Contents · Previous section · Next section

Section 24 of 36

24. Challenge uneven plate chemistry

First-round wells can differ in digestion, adaptor ligation or nuclei number; second-round wells can differ in amplification. Visualise every quality metric by row, column and processing date. A cluster enriched in one well pattern may be a batch signature rather than a biological cell state.

Contents · Previous section · Next section

Section 25 of 36

25. Challenge karyotype interpretations

Abnormal chromosome coverage may indicate aneuploidy, but it can also reflect sparse sampling, mapping bias or doublets. Require consistent dosage across bins, adequate contacts and replication across cells. Where possible, compare with cytogenetics, DNA sequencing or known cell-line karyotypes.

Contents · Previous section · Next section

Section 26 of 36

26. Keep specimens above cells in the hierarchy

Thousands of nuclei from one organism are repeated measurements, not thousands of independent biological replicates. For treatment, tissue or person-level claims, analyse cells within independent specimens and show whether the result recurs. This prevents a large cell count from inflating confidence.

Contents · Previous section · Next section

Section 27 of 36

27. Validate the decisive feature independently

Use conventional or single-cell Hi-C, imaging, DNA-content assays, sequencing or perturbation according to the claim. A barcode-derived trajectory gains credibility when an orthogonal measurement agrees; a structural contact gains credibility when it recurs across library designs or spatial imaging.

Contents · Previous section · Next section

Section 28 of 36

28. Build Primary Science process skills

Place coloured tokens into first-round cups, pool them, then redistribute them into second-round cups. The pair of colours identifies an origin until two tokens acquire the same pair. Students can vary cup number while holding token count constant, predict collisions and explain which design separates identities best.

Contents · Previous section · Next section

Section 29 of 36

29. Connect the PSLE Science habit of checking variables

Students can identify manipulated variables such as well count, responding variables such as collision frequency and controlled variables such as token number. Asking whether the evidence supports the conclusion mirrors PSLE Science reasoning while introducing the joyful idea that labels can scale experiments.

Contents · Previous section · Next section

Section 30 of 36

30. Extend the model in Secondary Science

Secondary learners can calculate the nominal combination space and compare it with observed usable combinations. They can graph occupancy, identify outliers and discuss why random allocation produces uncertainty. Biology, chemistry, mathematics and data literacy meet in one approachable indexing problem.

Contents · Previous section · Next section

Section 31 of 36

31. Use O-Level Science as a bridge

The official 2026 Singapore–Cambridge O-Level Biology and Chemistry syllabuses emphasise experimental design, measurement, data handling and evaluation. sci-Hi-C offers a current example: chemistry creates junctions, biology defines nuclei, and statistics tests whether indexed evidence supports a cell-level claim.

Contents · Previous section · Next section

Section 32 of 36

32. Ask school-choice questions that reveal fit

Families comparing science programmes can ask how students learn experimental design, coding, microscopy, genomics ethics and research communication. Confirm current subjects, enrichment and entry arrangements with each school’s official information. One advanced technique should inspire questions, not become an invented ranking of schools.

Contents · Previous section · Next section

Section 33 of 36

33. See the career pathways inside one barcode

The workflow involves molecular biologists, chemists, laboratory technologists, sequencing specialists, software engineers, statisticians, research coordinators and data stewards. Students do not need to choose immediately. The transferable skill is tracing how a claim moves from specimen to molecule, barcode, computation and decision.

Contents · Previous section · Next section

Section 34 of 36

34. Use science tuition as deliberate practice

Helpful science tuition can rehearse variable tables, calculation checks, graph reading and claim-evidence-reasoning with unfamiliar contexts such as sci-Hi-C. It should support school learning, not promise access to specific programmes or careers. Ask for visible corrections and explanations rather than keyword memorisation.

Contents · Previous section · Next section

Section 35 of 36

35. Did you know scale can increase the need for care?

More cells reveal rare patterns, but they also create more opportunities for collisions, batch effects and selective thresholds. That is good news for learners: scientific progress is not just bigger machines. It is better labels, clearer controls, auditable pipelines and honest uncertainty.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a publication checklist

Name the specimens, indexing scheme, collision model, plate map, read structure, per-cell filters, unique contacts, batch effects, biological hierarchy and independent validation. Share raw indexes, pairs, metadata, code and exclusions under appropriate governance. A trustworthy sci-Hi-C story makes every cellular identity traceable.

A reproducible sci-Hi-C project begins with an indexing budget rather than a sequencing budget. Write down the number of independent specimens, expected nuclei, first- and second-round wells, usable index sequences, anticipated recovery and acceptable collision rate. Simulate occupancy under uneven wells as well as ideal allocation. The protocol should state which index is attached at each molecular step and how the combination is read. This turns a large barcode space from an impressive number into a testable design.

Plate quality should be reviewed before biological labels are revealed. Plot reads, valid pairs, duplicates, cis fraction, contact distance and detected chromosomes by first-round well, second-round well, row, column and processing date. A diagonal or edge pattern suggests liquid-handling or temperature effects. If one cell cluster is concentrated in a plate region, repeat the analysis after removing that region. A real biological state may coexist with batch, but the evidence should show they can be distinguished.

Species mixing is especially informative when compatible materials are available. Nuclei from two species are combined, processed together and classified by mapped reads. A barcode containing substantial reads from both genomes suggests a collision or doublet. The test does not capture every same-species collision, so it estimates a component of error rather than certifying every cell. Report thresholds and the expected false-negative limitation. Synthetic barcode controls and unused combinations can separately measure index cross-talk.

Sparse contacts should be treated as samples from a much larger, unobserved set of possible interactions. Publish per-cell pair files and raw coverage before any imputation. When constructing embeddings or trajectories, repeat the analysis after matching contact counts, removing the richest cells and varying neighbourhood size. A stable biological organisation should not be entirely determined by library depth. If imputed contacts are required for a result, label that dependency in the figure and provide the unfilled comparison.

Cell-cycle analysis needs independent anchors. DNA content, replication timing, known chromosome-compaction patterns or held-out contact features can provide validation. The features used to order cells should not also serve as the only proof that the order is correct. Show contact quality and plate variables across the trajectory, because gradual technical drift can resemble progression. For karyotype inference, require broad, consistent chromosome dosage and compare with sequencing or cytogenetic information where possible.

Statistical independence must follow the biological question. Cells are nested within preparations and specimens; they are not interchangeable replicates for a treatment or donor claim. Use specimen-level summaries, mixed models or resampling that respects that hierarchy. Display how many cells and unique contacts each specimen contributes to each group. A rare state found in many cells from one preparation is interesting, but its generality remains an open question until it recurs independently.

Figures should begin with the molecular evidence funnel: nuclei quality, digestion, first-index balance, ligation, second-index balance, valid pairs, duplicates, collisions and retained cells. Follow with coverage-matched cell embeddings and only then show chromosome or contact interpretations. Every map needs bin size, colour scale, contact count and whether it represents one cell, an aggregate or an imputed matrix. Attractive structure should never hide a sparse or mixed barcode.

Data stewardship includes specimen provenance, consent where applicable, fixation and enzyme lots, plate maps, index sequences, raw index qualities, FASTQ files, alignments, pair files, per-cell metrics, exclusion tables, embeddings, code and computing environments. Human contact data can expose sequence and structural information and may require controlled access. Laboratory safety covers biological material, fixatives, enzymes, sorting, heat and amplified DNA. Accurate metadata are part of both safety and scientific reproducibility.

The next experiment should attack the largest uncertainty. Increase barcode diversity or reduce nuclei loading when collisions rise; repair liquid handling when plate effects appear; collect more independent specimens when one dominates; deepen libraries only when complexity curves justify it; and use imaging or another conformation assay when a spatial claim is central. State the result that would support, narrow or contradict the proposed model before collecting the follow-up data.

For students and parents, a useful reading routine has four columns: physical action, recorded value, computational inference and remaining alternative. sci-Hi-C physically indexes and ligates nuclei, records read pairs and barcodes, infers cellular contact patterns and must still consider collisions and sparse sampling. This routine makes an advanced genomics paper approachable while reinforcing fair-test reasoning from Primary Science through O-Level Science.

A pre-publication barcode audit can be performed blind to condition. An analyst receives index sequences, qualities, molecule counts, species assignments, contact metrics and plate positions, then fixes demultiplexing and filtering rules before seeing biological groups. After labels are revealed, any disproportionate losses are investigated rather than explained away. This modest separation reduces selective thresholding and creates a durable record of why each cell identity was accepted.

Finally, publish a compact sensitivity table. Rows can vary mismatch allowance, minimum contacts, collision threshold, doublet rule, contact downsampling and specimen inclusion; columns can show whether each headline conclusion survives. Stability does not require identical numbers, but the direction and scope of the claim should remain recognisable. When it does not, the narrower conclusion is the scientifically useful one. Scale becomes meaningful when every extra cell strengthens an auditable inference rather than merely enlarging a total.

Contents · Previous section · Continue to the Science Learning Hub