Experimental Models and Screens
TL;DR
Cancer models are simplified systems for asking causal questions. 2D cell lines are fast and scalable. 3D spheroids and organoids better preserve architecture and some patient-specific biology. PDX and GEMM models add organism-level context. CRISPR screens perturb genes across many cells to discover dependencies, resistance mechanisms, and synthetic lethal relationships. Sources: [1], [2]
Model comparison
| Model | Best for | Major limitation |
|---|---|---|
| 2D cell line | scalable perturbation and drug screens | adaptation to plastic |
| 3D spheroid | gradients, penetration, hypoxia | simplified architecture |
| organoid | patient-derived epithelial biology | variable immune/stromal preservation |
| co-culture | cell-cell interaction | design complexity |
| PDX | in vivo tumor growth and drug response | mouse context, cost, time |
| GEMM | tumor initiation and immune context | model-specific biology |
| CRISPR screen | gene-function discovery | artifacts, copy-number bias, fitness confounding |
CRISPR screen logic
- Deliver sgRNA library.
- Select or treat cells.
- Sequence sgRNA abundance.
- Compare depletion/enrichment.
- Correct artifacts.
- Validate top hits orthogonally.
CRISPR screens are powerful because they convert gene function into a measurable population shift. They are dangerous when treated as direct patient truth without validation.
What each model can answer
| Question | Better fit |
|---|---|
| Is this gene essential in KRAS-mutant cells? | CRISPR screen |
| Does this drug penetrate a 3D mass? | spheroid/organoid |
| Does this patient's tumor respond ex vivo? | patient-derived organoid |
| Does therapy affect whole-organism toxicity? | animal model |
| Does immune context matter? | co-culture, humanized model, GEMM |
Computational output
- viability curves
- dose-response matrices
- IC50/AUC
- growth-rate adjusted metrics
- sgRNA count matrices
- MAGeCK/Chronos/BAGEL-style dependency scores
- model metadata
- histology and imaging features
Developer notes
- Model identity, passage number, culture medium, and mycoplasma status are not optional metadata.
- Drug-response curves should preserve raw viability, normalization, exposure time, and growth rate.
- CRISPR screen outputs need library, guide sequence, copy-number correction, and replicate metadata.
- A dependency score is a prioritization signal, not a validated drug target.
- Organoid response can reflect culture conditions as much as patient biology.
- Always separate discovery screens from validation assays in the data model.
References
- Tong L, Cui W, Zhang B, et al. Patient-derived organoids in precision cancer medicine. Med 2024;5:1351-1377. PMID 39341206. https://doi.org/10.1016/j.medj.2024.08.010
- Behan FM, Iorio F, Picco G, et al. Prioritization of cancer therapeutic targets using CRISPR-Cas9 screens. Nature 2019;568:511-516. PMID 30971826. https://doi.org/10.1038/s41586-019-1103-9