Perturbation Lab · bounded preview
Connect a perturbation to response programs—without inventing a counterfactual.
Inspect observed HepG2 CRISPRi evidence, compare two candidate genes under an explicit additive assumption, and anchor each candidate to a real human-myogenic 0/24/48/72-hour reference while keeping unsupported knockout, overexpression and perturbed-fate outputs locked.
Perturbation evidence and preview ladder
Every view carries its evidence role
Capability boundary
Available now versus data required
Evidence-backed views
- Observed CRISPRi · 21 frozen HepG2 singleton aggregates
- Additive hypothesis · δA + δB null preview with no interaction term
- Reference trajectory · real unperturbed human-myogenic time course
Locked outputs
- Knockout · matched KO and editing calibration
- CRISPRa / overexpression · matched muscle or DMD activation response
- Interaction · observed same-screen doubles
- Perturbed fate · dynamic perturbation sampling
Locked items expose only their unlock conditions; they do not emit simulated values.
External evidence registry
Four public reference datasets are qualified—each with a different permitted role
The cards separate direct perturbation evidence, muscle phenotype screening, DMD correction, unperturbed myogenic time evolution and cross-cell-line CRISPRa-combination calibration.
Current substrateHepG2 CRISPRi
21 observed single perturbations remain the only direct response substrate in the browser.
- Direct candidates
- 21
- Observed doubles
- 0
GSE293514Human myoblast fusion screen
Muscle-context KO/fusion qualification; absence from the fusion-hit set is endpoint-specific, not proof of no muscle function.
- 21-gene library coverage
- 9/21
- Fusion hits
- 0
GSE272233DMD CRISPR-correction reference
Bulk RNA-seq from three exon-duplication backgrounds and matched corrected clones; candidate transcripts are observations, not candidate perturbation effects.
- Samples
- 21
- Correction DE calls
- 5923
GSE52529Human myogenic time course
Primary human skeletal myoblast single cells at 0, 24, 48 and 72 hours, with distributed pseudotime and state assignments.
- Cells
- 271
- Candidate coverage
- 20/21
GSE133344CRISPRa + pair calibration
K562 single and combinatorial CRISPRa benchmark. It calibrates method design, not muscle-context candidate overexpression effects.
- Reported perturbations
- 287
- Feature coverage
- 20/21
SIM-L0 · observed CRISPRi
ZNF133 observed CRISPRi
Muscle-context assay. This is an observed HepG2 CRISPRi record linked to descriptive program inputs; independent muscle/DMD perturbation replication is absent.
Observed cells84Frozen HepG2 CRISPRi aggregate
Counteralignment0.064Contextual signature relationship; not therapeutic rescue
Skeletal-muscle expression4.56 TPMExpression context, not perturbation validation
Independent DMD replication0Scientific unlock remains unmet
Selected-gene external evidence
Context, correction, time-course and calibration evidence
GSE293514 · human myoblast KOAssessed
Fusion endpoint context only; not DMD rescue.
GSE272233 · DMD correction0/3 significant contrasts
ZNF133 transcript response after DMD editing, not ZNF133 perturbation.
GSE52529 · myogenic time courseMeasured reference
Unperturbed 0/24/48/72-hour reference; perturbed fate remains locked.
GSE133344 · CRISPRa calibrationExpression feature present
Feature presence is not target coverage and does not unlock overexpression.
Pathway / program lens
Descriptive overlap with frozen gene programs
Myogenic / contractile0
No compact-list overlap
ECM / fibrotic2
CTHRC1, SPP1
Inflammatory / stress1
SOD2
Cell cycle / proliferation2
MKI67, STMN1
Metabolic / secretory3
APOA1, GPX3, MLXIPL
Proteostasis / membrane1
GPX3
Response details
Compact response-gene lists
APOA1CTHRC1MKI67STMN1CDT1CTNNAL1SERPINE1SOD2MT2ASPP1MLXIPLGPX3
Server-rendered frozen inputs; descriptive only.Current boundary
Observed same-HepG2 CRISPRi evidence only; independent DMD validation is absent.
Open ZNF133 evidence record · Open Study CardUnlock conditions
What the next data release must add
Editing efficiency, residual expression, controls and context-matched outcomes; CRISPRi cannot be relabelled as KO.
GSE133344 now calibrates study and pair design, but activation dose and expression response must be observed in muscle/DMD context; the negative of a knockdown response is not accepted.
Same-screen singletons and doubles with pair-disjoint validation before estimating γAB, synergy or synthetic rescue.
GSE52529 supplies the unperturbed 0/24/48/72-hour reference. Candidate-conditioned velocity or fate still requires matched perturbation cells sampled across the lineage.
Boundary. The program lens is a descriptive overlap display, not formal pathway enrichment. The pair view is an additive hypothesis, not an observed double perturbation. No output is a treatment, safety, causal-mechanism or clinical recommendation.