Why Science? | SCENIC, Regulons, Cis-Regulatory Motifs and Cell-State Evidence

Section 1 of 36

1. Start with the regulon question

A cell’s RNA profile contains thousands of genes, but SCENIC asks a more organised question: which transcription factors may coordinate coherent target programmes in this cell? The answer is expressed as regulons and their activity. That is stronger than a list of markers, yet still a model built from expression and motif evidence.

Contents · Next section

Section 2 of 36

2. Build co-expression modules

The first stage searches for genes whose expression covaries with a transcription factor across cells. The original workflow used tree-based network inference such as GENIE3; scalable implementations can use alternatives such as GRNBoost. Co-expression supplies candidates. It does not distinguish direct regulation from shared cell state, technical variation or an unmeasured common cause.

Contents · Previous section · Next section

Section 3 of 36

3. Add cis-regulatory motif evidence

SCENIC then asks whether candidate targets contain enriched motifs compatible with the transcription factor. This cis-regulatory evidence helps prune indirect co-expression relationships and define a regulon. Motif presence is opportunity, not observed occupancy: the site may be inaccessible, bound by another factor or inactive in the biological context.

Contents · Previous section · Next section

Section 4 of 36

4. Understand a regulon

A regulon is a transcription factor paired with a supported set of candidate targets. Regulons can overlap because biological programmes share genes and factors cooperate. The useful unit is therefore a documented target set with provenance, not a slogan that one factor controls an entire cell identity.

Contents · Previous section · Next section

Section 5 of 36

5. Score activity with AUCell

AUCell ranks genes within each cell and asks whether a regulon’s targets are enriched near the top of that ranking. This within-cell strategy can be less dependent on global expression scaling than a raw average. The score reports recovery of a target set, not transcription-factor protein abundance or binding strength.

Contents · Previous section · Next section

Section 6 of 36

6. Identify cell states

Regulon activity profiles can cluster cells by regulatory programme and sometimes reveal states that ordinary gene-expression clustering obscures. Interpret a new state only when it is reproducible across specimens and supported by markers, anatomy or function. A clean activity heat map does not create a biological category by itself.

Contents · Previous section · Next section

Section 7 of 36

7. Keep the 2017 boundary

The primary Nature Methods paper reported simultaneous network reconstruction and cell-state identification and demonstrated the workflow in single-cell datasets including tumour and brain examples. Later repositories and protocols extend implementation and scale. Cite the paper for evaluated claims and the exact software revision for the pipeline actually run.

Contents · Previous section · Next section

Section 8 of 36

8. Create a specimen manifest

Record organism, tissue, condition, donor, protocol, batch, cell identifiers and biological replicate. Cells are nested within specimens. Thousands of droplets from one donor do not provide thousands of independent biological replicates, so network stability and group comparisons must be summarised at an appropriate experimental unit.

Contents · Previous section · Next section

Section 9 of 36

9. Inspect expression quality

Filter low-quality cells, doublets and genes using rules chosen before inspecting headline regulons. Plot library size, detected genes, mitochondrial fraction and stress programmes by cell type and specimen. A technical gradient can become a co-expression module and survive motif pruning if quality is correlated with biological annotation.

Contents · Previous section · Next section

Section 10 of 36

10. Choose the regulatory search space

Specify which transcription factors, genes and motif collections enter the analysis. Identifier mapping, species conversion and duplicated symbols can change target sets silently. Archive the motif database version and ranking files because the regulatory search space is part of the result, not a disposable implementation detail.

Contents · Previous section · Next section

Section 11 of 36

11. Protect rare populations

Rare cells may lack enough observations for stable co-expression estimates. Pooling them with a broad parent population may erase specific biology, while analysing them alone may create unstable edges. Report cell counts per specimen and use resampling to show whether a rare-state regulon recurs rather than celebrating one fragile network.

Contents · Previous section · Next section

Section 12 of 36

12. Control batch and composition

A factor can appear active because one batch contains more of the cell type in which its targets are expressed. Compare activity within cell type and specimen, visualise batches separately and avoid regressing away genuine condition biology without checking. Composition shifts and within-cell changes answer different scientific questions.

Contents · Previous section · Next section

Section 13 of 36

13. Pin the implementation

Record package versions, random seeds, network-inference settings, motif rankings, organism annotations and thresholds. SCENIC, pySCENIC and scalable protocols are related but not interchangeable labels. Reproducibility requires the exact route from counts to regulons rather than the broad method family name.

Contents · Previous section · Next section

Section 14 of 36

14. Separate discovery from confirmation

Use one cohort to discover regulons and another to score or confirm them when possible. Reconstructing and celebrating the same network in the same cells can overstate stability. External scoring, held-out specimens or split-sample analysis makes the evidence boundary visible.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with a fictional regulon table

This classroom table is invented and is not SCENIC output.

Contents · Previous section · Next section

Section 16 of 36

16. Read activity distributions

Show per-cell distributions and specimen summaries, not only a heat map average. A high mean can reflect a small subgroup or broad moderate activity. Report thresholds if activity is binarised and test whether conclusions persist without binarisation, because a convenient on/off display can conceal continuous biology.

Contents · Previous section · Next section

Section 17 of 36

17. Inspect the target genes

For every headline regulon, list leading targets and their direction, expression coverage and biological coherence. A familiar transcription-factor name can seduce readers into accepting an incoherent target set. Targets are the evidence-bearing layer and should be visible beside the activity score.

Contents · Previous section · Next section

Section 18 of 36

18. Compare expression and regulon views

Place transcription-factor RNA, target-gene expression and AUCell activity together. A factor can have modest RNA while its programme is active, or high RNA with little target recovery. These differences are scientifically interesting, but they do not prove post-transcriptional activation without protein or binding measurements.

Contents · Previous section · Next section

Section 19 of 36

19. Test cluster dependence

Recalculate summaries without relying on the final cluster labels. If a regulon exists only because the same expression features produced both the cluster and the activity claim, interpretation may be circular. Continuous embeddings, independent annotations and held-out labels can show whether the programme survives a different representation.

Contents · Previous section · Next section

Section 20 of 36

20. Compare with simple baselines

Benchmark SCENIC activity against average target expression, canonical markers and random gene sets matched for size and abundance. Complexity earns its place when motif-informed regulons are more stable, discriminating or experimentally useful than transparent baselines.

Contents · Previous section · Next section

Section 21 of 36

21. Quantify recurrence

Bootstrap cells within specimens, rerun network inference and count how often edges and regulons return. Also reconstruct networks after leaving out one specimen. Recurrence is more informative than one network’s edge weight because high-dimensional algorithms can select different but similarly scoring relationships.

Contents · Previous section · Next section

Section 22 of 36

22. Audit motif ambiguity

Related transcription factors often recognise similar DNA motifs. A motif-enriched target set may support a family more confidently than one precise factor. Preserve motif-to-factor mapping rules and use expression, accessibility or binding data to resolve family members rather than overclaiming specificity.

Contents · Previous section · Next section

Section 23 of 36

23. Audit chromatin context

Motif support does not ensure that a site is accessible in the studied cell. When chromatin-accessibility data exist, ask whether predicted sites are open in the relevant state. When they do not, label accessibility as an unmeasured link in the evidence chain.

Contents · Previous section · Next section

Section 24 of 36

24. Audit cell cycle and stress

Cell-cycle, interferon, heat-shock and dissociation programmes can dominate co-expression. Score them explicitly, rerun analyses after reasonable controls and show whether the regulatory conclusion remains. Removing every stress gene is not automatically correct; the aim is to distinguish biology from preparation artefact.

Contents · Previous section · Next section

Section 25 of 36

25. Audit network direction

SCENIC’s directed factor-to-target language is supported by motif knowledge, but the single-cell expression data remain observational. Feedback, co-regulation and time lag complicate direction. Use perturbation or time-resolved evidence before claiming that changing the factor will produce a particular target response.

Contents · Previous section · Next section

Section 26 of 36

26. Audit multiple discovery choices

Networks depend on gene filtering, algorithm, motif database, pruning and activity thresholds. A result selected after trying many settings needs a sensitivity record. Predeclare important comparisons where possible and label exploratory optimisation honestly.

Contents · Previous section · Next section

Section 27 of 36

27. Design a decisive experiment

Choose a regulon that recurs across specimens and has accessible candidate sites. Perturb the factor, measure target RNA, factor binding or chromatin change, include a related-factor control and test rescue. The experiment should discriminate the proposed regulator from a shared state programme.

Contents · Previous section · Next section

Section 28 of 36

28. Report an evidence ladder

Level one is co-expression, level two adds motif enrichment, level three adds cell-state recurrence, level four adds accessibility or binding, and level five adds perturbation with rescue. A concise ladder lets readers see how far the claim has travelled without flattening every result into ‘regulation’.

Contents · Previous section · Next section

Section 29 of 36

29. Connect to school science

Regulons turn textbook ideas about DNA, genes and proteins into an evidence-reading exercise. Students can ask what is measured, what is inferred and what experiment tests the link. That habit supports Secondary Science and O-Level Biology without requiring learners to run the software.

Contents · Previous section · Next section

Section 30 of 36

30. Build a paper activity

Give learners fictional expression ranks and a small target set, then calculate whether targets gather near the top. Add a motif card and a knockout result in stages. The class can watch a weak correlation become a stronger, still bounded regulatory claim.

Contents · Previous section · Next section

Section 31 of 36

31. Connect to computing

The workflow combines ranking, tree-based inference, enrichment and reproducible pipelines. It offers a friendly route from biology to algorithms: every computational choice changes an evidence object. Students can discuss data structures, random seeds, scalability and why validation is not replaced by faster code.

Contents · Previous section · Next section

Section 32 of 36

32. Connect to mathematics

Rank enrichment, resampling and similarity connect the article to probability and statistics. Ask how target-set size changes an enrichment score, why matched random sets matter and what an uncertainty interval communicates. Mathematics becomes a way to control enthusiasm rather than a decorative calculation.

Contents · Previous section · Next section

Section 33 of 36

33. Connect to career pathways

Gene-regulatory analysis appears in genomics, developmental biology, cancer research, bioinformatics and biotechnology. The honest career message is about transferable practices—coding, experimental design, statistics, documentation and teamwork—not a guaranteed role or admission outcome from learning one package.

Contents · Previous section · Next section

Section 34 of 36

34. Use AI as a checked assistant

AI can explain code, propose plots and help convert an evidence ladder into plain language. It must not invent target genes, motif sources, software options or citations. Verify every method claim against the primary paper and official documentation, and keep generated prose separate from measured results.

Contents · Previous section · Next section

Section 35 of 36

35. Run the publication checklist

Confirm title and slug uniqueness, source links, software and motif versions, specimen counts, fictional labels, table semantics, internal owner pages and claim boundaries. Search for phrases such as ‘directly controls’ and replace them unless binding and perturbation evidence justify that wording.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with the right claim

SCENIC can organise single-cell expression into testable regulatory programmes and illuminate cell states. Its best output is not a final causal map; it is a transparent shortlist of regulons, targets and experiments whose evidence can be strengthened one layer at a time.

A careful SCENIC report should make three objects downloadable: the factor-to-target network before motif pruning, the final regulons after pruning, and the cell-by-regulon activity matrix. Keeping all three prevents a polished heat map from becoming detached from the evidence that created it. It also allows another analyst to ask whether a conclusion depends on the network stage or the activity stage.

Did you know? Two cells can express a transcription factor at similar RNA levels while receiving different regulon-activity scores. The difference may come from target-gene recovery, cofactors, chromatin state or technical coverage. That pattern is an excellent hypothesis generator, but only protein, binding, accessibility and perturbation evidence can decide which explanation is correct.

Use sensitivity panels as part of the main story. Rebuild networks with at least one defensible alternative inference setting, vary motif-pruning thresholds and show the overlap of leading regulons. Then rescore a held-out dataset using the discovered regulons. Stability across these steps is more persuasive than a single beautifully ordered heat map.

Negative evidence needs gentle language. If a regulon does not score in a cell type, write that the available data and target set do not support activity. Sparse counts, missing motif annotations or a context-specific target programme can all produce a low score. ‘Inactive’ is a biochemical conclusion that requires stronger measurement.

For families deciding whether a child enjoys science, the lesson is wonderfully accessible: scientists rarely see a mechanism all at once. They combine patterns, prior knowledge and tests. A student who likes asking ‘what evidence would change my mind?’ is already practising the heart of regulatory genomics.

One useful figure pairs a regulon network with a coverage strip. The network shows factor-to-target relationships, while the strip shows how often each target was detected in every specimen. This prevents dense network artwork from hiding weak observation. Add a motif-family annotation so readers can see when factor identity is precise and when the sequence evidence supports only a family.

Regulon comparisons across conditions should preserve the same target definition when the question is activity change. Reconstructing a separate network in each condition asks a different question because both edges and scores can move. Publish both only when their meanings are clearly separated: a fixed regulon tests activity, while independently inferred networks explore rewiring.

A responsible conclusion can be exciting and narrow at once: ‘These cells consistently recover the targets of this motif-supported regulon, and the programme is a strong candidate for perturbation.’ That sentence invites an experiment without pretending the experiment has occurred. It also tells students that scientific confidence grows through specificity, not through dramatic verbs.

Finally, archive the failed checks. A regulon that disappeared under leave-one-donor-out analysis, a motif family that could not resolve one factor, or a target set that matched random controls is valuable information. Publishing those boundaries improves later study design and prevents another team from mistaking the same fragile pattern for a firm mechanism.

A final reproducibility test is deliberately simple: another analyst should be able to take the saved regulon table and activity matrix and reproduce every figure without rerunning network inference. Then, with the documented software environment and motif rankings, that analyst should also be able to rebuild the network. Separating figure reproduction from full pipeline reproduction makes failures easier to locate and gives students a concrete picture of open, checkable science.

The science remains joyful because every uncertainty suggests a useful question. Does chromatin open before the regulon rises? Does a related factor recognise the same motif? Does perturbation remove the target programme, and can rescue restore it? SCENIC helps arrange those questions in a productive order while the laboratory decides which regulatory story survives.

Contents · Previous section · Continue to the Science Learning Hub