eduKateSG · Why Science?
Ask whether a guide RNA produced a detectable molecular response in each cell—and keep classification separate from mechanism
Reading routes
Science learning becomes useful when a familiar object or observation is turned into a system of quantities, mechanisms and claim limits. This guide owns one applied evidence-reading job inside eduKateSG’s wider Science estate. It connects naturally to Why Science Gears Knowledge Graphs Multigene Perturbation Prediction; Why Science Perturbnet Unseen Chemical Structures Genetic Perturbation Responses; Why Science Cinema Ot Confounder Separation Single Cell Treatment Effects; Why Science Cpa Compositional Perturbations Dose Response Prediction; Education Hub; Singapore Secondary School Directory; Career Adulthood Hub; Science Learning Hub. It also keeps current school and public claims traceable to visible primary sources: Mixscape primary study; Mixscape PubMed record; Official Seurat Mixscape vignette; Official RunMixscape reference; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science syllabus; 2026 MOE G2 Computing syllabus. The sources describe the scientific scope; this article translates that scope into a calm route for Primary Science, PSLE Science, Secondary Science, O-Level Science, STEM exploration, school choices and career pathways without inventing admission or employment outcomes.
Mixscape, introduced by Papalexi and colleagues in Nature Genetics in 2021, addresses a practical complication in pooled single-cell CRISPR screens: carrying a guide RNA does not guarantee that a cell shows the intended molecular response. The method estimates perturbation signatures and classifies cells as perturbed or non-perturbed relative to controls, helping separate a detectable response from background variation. This can sharpen downstream analysis, but a classification is not proof of editing efficiency, target engagement, mechanism or clinical value.
Inside this guide
1–12 · Foundations and models
- 1. Begin with guide-positive uncertainty
- 2. Separate assignment from response
- 3. Use non-targeting controls
- 4. Estimate target-specific signatures
- 5. Classify cells carefully
- 6. Keep multimodal evidence distinct
- 7. Read the 2021 evidence in scope
- 8. Define the experimental unit
- 9. Create a guide manifest
- 10. Inspect guide multiplicity
- 11. Freeze quality control
- 12. Model cellular context
13–24 · Evidence, testing and applications
- 13. Check the perturbation-signature direction
- 14. Use replicate-aware validation
- 15. Practise with a fictional Mixscape table
- 16. Read a perturbed class proportion
- 17. Read a non-perturbed class
- 18. Inspect marker changes
- 19. Check surface-protein evidence
- 20. Audit parameter sensitivity
- 21. Run negative controls
- 22. Validate editing directly
- 23. Compare guide concordance
- 24. Test downstream robustness
25–36 · Learning, decisions and pathways
- 25. Compare with GEARS
- 26. Compare with CINEMA-OT
- 27. Build a perturbation evidence card
- 28. Design the next experiment
- 29. Connect Mixscape to school science
- 30. Try a signal-versus-label activity
- 31. Connect to mathematics
- 32. Connect to computing and AI literacy
- 33. Connect to school and career choices
- 34. Create a family evidence habit
- 35. Use precise Mixscape language
- 36. Finish with a guide-to-validation bundle
Section 1 of 36
1. Begin with guide-positive uncertainty
Pooled CRISPR assays attach guide identities to cells, yet delivery and guide detection do not guarantee a functional perturbation. The scientific question begins after assignment.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 2 of 36
2. Separate assignment from response
A barcode says which perturbation was intended. A response signature asks whether the cell’s measured molecular state differs consistently from suitable controls.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 3 of 36
3. Use non-targeting controls
Control guides estimate background variation associated with the experiment, delivery system and assay. They provide a reference rather than a perfectly inert population.
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 4 of 36
4. Estimate target-specific signatures
Mixscape compares target-associated cells with controls while accounting for local cellular context, producing a perturbation signature for classification and analysis.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 5 of 36
5. Classify cells carefully
Cells can be labelled as perturbed or non-perturbed within a target group according to the inferred response. The labels summarise evidence under the fitted model.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 6 of 36
6. Keep multimodal evidence distinct
RNA, surface protein, guide capture and editing-genotype assays measure different layers. Agreement strengthens a story; disagreement can expose timing or mechanism questions.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Section 7 of 36
7. Read the 2021 evidence in scope
Papalexi and colleagues introduced Mixscape for pooled single-cell CRISPR screening and demonstrated analyses that separate perturbation responses from confounding variation.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 8 of 36
8. Define the experimental unit
A cell is an observation, but the replicate is usually a culture, donor, animal or experimental run. Thousands of cells do not replace independent biology.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 9 of 36
9. Create a guide manifest
Record target, guide sequence, vector, capture method, intended edit, multiplicity, exposure time, cell system, batch and replicate before modelling.
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 10 of 36
10. Inspect guide multiplicity
Cells carrying multiple guides can blur target-specific signatures. State how multiplets, ambiguous assignments and low guide counts were handled.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 11 of 36
11. Freeze quality control
Choose mitochondrial, feature, count and doublet rules without looking for a desired target result. Post-hoc filtering can manufacture cleaner separation.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 12 of 36
12. Model cellular context
A response may differ across cell states. Compare target cells with relevant controls rather than treating every cell in a heterogeneous sample as exchangeable.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Section 13 of 36
13. Check the perturbation-signature direction
Inspect which genes or proteins drive separation and whether they match plausible downstream biology, while avoiding circular selection and evaluation.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 14 of 36
14. Use replicate-aware validation
Learn or tune on some experiments and test on untouched experiments. Randomly holding out cells from the same culture is an easier question.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 15 of 36
15. Practise with a fictional Mixscape table
This classroom table is invented and is not output from the Mixscape paper.
| Fictional target | Guide-positive cells | Perturbed class | Genotype check | Next action |
|---|---|---|---|---|
| Gene A | 900 | 71% | Concordant | Replicate |
| Gene B | 820 | 34% | Mixed | Sequence |
| Gene C | 760 | 9% | Not run | Measure first |
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 16 of 36
16. Read a perturbed class proportion
The fraction classified as perturbed describes detectable response under chosen settings. It is not identical to editing efficiency measured at DNA or protein level.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 17 of 36
17. Read a non-perturbed class
Non-perturbed cells are scientifically informative. They can reveal technical failure, biological resistance, delayed response or an assay unable to see the intended consequence.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 18 of 36
18. Inspect marker changes
Downstream differential analysis should compare clearly defined populations and disclose that class selection preceded testing. Selection and testing on the same signal can inflate confidence.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Section 19 of 36
19. Check surface-protein evidence
When protein measurements exist, compare them with RNA signatures and direct target abundance. Different half-lives can create sensible delays between modalities.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 20 of 36
20. Audit parameter sensitivity
Rerun neighbourhood, dimensionality and classification choices. Stable conclusions should not depend on one undocumented setting.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 21 of 36
21. Run negative controls
Non-targeting guides, irrelevant targets and label permutations reveal whether the pipeline invents target structure from batch or cell-state imbalance.
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 22 of 36
22. Validate editing directly
Amplicon sequencing, protein measurement or another target-specific assay can test whether the intended edit occurred. These measurements answer a different question from expression classification.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 23 of 36
23. Compare guide concordance
Independent guides aimed at the same gene should support a shared biological pattern. Guide-specific divergence can indicate off-target effects or efficacy differences.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 24 of 36
24. Test downstream robustness
Repeat pathway, interaction or cluster analyses with all guide-positive cells and with Mixscape classes. Show which conclusions depend on the filter.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Section 25 of 36
25. Compare with GEARS
GEARS predicts responses to genetic perturbations, including combinations; Mixscape asks whether a measured guide-positive cell exhibits a detectable perturbation signature.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 26 of 36
26. Compare with CINEMA-OT
CINEMA-OT estimates treatment effects inside measured contrasts using confounder separation and matching. Mixscape focuses on response classification in pooled perturbation screens.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 27 of 36
27. Build a perturbation evidence card
Place guide assignment, class probability, replicate consistency, direct editing evidence, modality agreement and claim wording on one page.
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 28 of 36
28. Design the next experiment
Choose independent guides, direct genotype or protein validation, untouched replicates and a time course that distinguishes absent from delayed responses.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 29 of 36
29. Connect Mixscape to school science
Variables, controls, classification and fair tests become concrete when one intended intervention does not always produce the same observable outcome.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 30 of 36
30. Try a signal-versus-label activity
Give learners labelled cards with overlapping measurements, let them set a decision rule, then reveal direct validation and discuss false positives and false negatives.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Section 31 of 36
31. Connect to mathematics
Distances, probability distributions, classification thresholds, confusion matrices and uncertainty turn a qualitative response into testable quantitative evidence.
In a pooled screen, guide identity is an assignment; the measured expression profile is the evidence for whether a detectable response followed.
Archive guide sequences, target annotations, control guides, multiplicity, time, batches, preprocessing, parameters, seeds and Seurat version.
Section 32 of 36
32. Connect to computing and AI literacy
A model can organise noisy observations, but its classes inherit the controls, features and thresholds chosen by people.
Mixscape estimates a perturbation signature relative to matched non-targeting controls and uses that signature to distinguish perturbed from non-perturbed cells.
Keep biological replicates intact and determine quality-control thresholds before inspecting attractive target-specific patterns.
Section 33 of 36
33. Connect to school and career choices
Students who enjoy experiments and data may explore biology, genetics, statistics, computing, bioinformatics or laboratory technology without assuming one fixed pathway.
A non-perturbed classification can reflect incomplete editing, weak downstream transcription, timing, measurement noise or biological compensation.
Compare guide assignment alone, all guide-positive cells, simple control-distance rules and Mixscape classes on the same held-out experiments.
Section 34 of 36
34. Create a family evidence habit
When a headline says a gene was switched off, ask what was delivered, what response was measured and whether the edit itself was checked.
The method improves signal-to-noise for downstream comparisons, but filtering changes the population being described and must be reported.
Report class proportions, signature strength, uncertainty, replicate consistency, genotype checks and downstream sensitivity to filtering.
Section 35 of 36
35. Use precise Mixscape language
Say guide-positive cell, inferred perturbation signature and perturbed or non-perturbed class. Reserve successful editing and mechanism for direct supporting evidence.
Papalexi and colleagues combined transcript and surface-protein measurements in their study, illustrating how multimodal evidence can refine interpretation.
Build a perturbation card with intended target, measured modality, controls, class rule, validation status and unresolved alternatives.
Section 36 of 36
36. Finish with a guide-to-validation bundle
Deliver manifests, control definitions, replicate-aware splits, class diagnostics, direct assays, sensitivity analyses, evidence cards and the next discriminating experiment.
Perturbation class, editing genotype and causal mechanism are different evidence layers; a rigorous report never swaps one for another.
Say detectable response signature or Mixscape perturbation class; do not say successful edit, target engagement or mechanism without direct evidence.
Did you know? A guide-positive cell can carry the intended barcode yet show no detectable downstream expression response. Mixscape was built around that gap between assignment and response.
The final report should keep three columns separate: intended perturbation, inferred response class and direct editing validation. That simple structure prevents an algorithmic label from silently becoming a molecular fact.
For human-derived data, preserve consent-aware access controls and donor-level validation; reproducibility does not require unrestricted release of identifiable raw profiles.
Contents · Previous section · Continue to the Science Learning Hub