# Lab models that fail to predict what happens in patients

Source: https://onco.cc/bottlenecks/b-preclinical-models/  
OnCo record `b-preclinical-models` (Bottleneck). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Nine in ten cancer drugs that work in mice fail in humans. Our models are the reason.

## Summary

Oncology has the lowest probability of success of any therapeutic area: only a few percent of agents entering phase 1 reach approval, and most failures are for lack of efficacy that the preclinical package did not anticipate. Immortalised cell lines have drifted for decades and lack a microenvironment; subcutaneous xenografts grow in immunodeficient mice with no human immune system, stroma or metastatic pattern; genetically engineered mouse tumours evolve with far less heterogeneity than human disease; and patient-derived organoids capture epithelial biology but not vessels, immune cells or drug pharmacokinetics. Preclinical studies are also small, unblinded and rarely replicated. The result is a pipeline that spends billions in humans to learn what the models could not tell it. Better-validated models, functional testing on fresh patient tissue, and systematic benchmarking of model predictions against clinical outcomes are the fixes.

## Fields

- Kind: Bottleneck
- Last checked: 2026-09-08
- Stage: biology
- Severity: critical
- Metrics: Probability that an oncology drug entering phase 1 is eventually approved: 3.4% (Wong, Siah & Lo, Biostatistics 2019); Oncology likelihood of approval from phase 1 (industry data 2003-2011), lowest of all disease areas: 6.7% (Hay et al., Nature Biotechnology 2014); Landmark preclinical cancer papers whose findings Amgen scientists could reproduce: 6 of 53 (11%) (Begley & Ellis, Nature 2012)
- Causes: Cell lines have adapted to plastic for decades and no longer resemble the tumours they came from.; Xenografts require immunodeficient hosts, so anything involving the immune system is invisible.; Mouse tumours are clonally simpler and are treated when small, which overstates drug effect.; Organoids lack stroma, vasculature and immune cells, and organoid drug exposure is not human pharmacokinetics.; Preclinical efficacy studies are small, unblinded, unregistered and rarely include controls that match clinical practice.; There is no systematic scoring of which models predicted which clinical results, so the field cannot learn which models to trust.

## Sources

- Wong, Siah & Lo, Estimation of clinical trial success rates (Biostatistics 2019): https://doi.org/10.1093/biostatistics/kxx069
- Begley & Ellis, Raise standards for preclinical cancer research (Nature 2012): https://doi.org/10.1038/483531a
- Cancer Models / PDCM Finder: https://www.cancermodels.org/

## Connected records

- collections: [Cancer Models (PDCM Finder) & HCMI](https://onco.cc/collections/cancer-models/), [DepMap (Cancer Dependency Map)](https://onco.cc/collections/depmap/)
- ideas: [A bone marrow niche on a chip to study human dormancy](https://onco.cc/ideas/idea-bio2-marrow-niche-on-chip/), [A digital passport for every cell culture: identity, contamination status, passage number](https://onco.cc/ideas/idea-tr2-cell-line-passport/), [A drug screen that only rewards killing sleeping cancer cells](https://onco.cc/ideas/idea-bio2-dormancy-selective-screen/), [A funded organoid and PDX panel as the go/no-go gate before IND-enabling money](https://onco.cc/ideas/idea-fund-organoid-translation-gate/), [A global rapid tissue donation network for metastatic disease](https://onco.cc/ideas/idea-bio2-rapid-autopsy-commons/), [A home for the animal and organoid experiments that failed](https://onco.cc/ideas/idea-bio1-negative-preclinical-repository/), [A phase 0 fund to test academic compounds in humans with microdoses and imaging](https://onco.cc/ideas/idea-fund-phase-zero-fund/), [A public atlas of drug-pair responses across a thousand patient-derived organoids](https://onco.cc/ideas/idea-tr2-organoid-matrix-atlas/), [A virtual cancer cell that predicts what a drug will do before you test it](https://onco.cc/ideas/idea-bio1-virtual-cell-perturbation/), [An open engine that ranks every drug pair by predicted synergy before anyone runs a trial](https://onco.cc/ideas/idea-tr2-synergy-ranking-engine/), [An open foundation model of the cancer cell trained on perturbation data](https://onco.cc/ideas/idea-data-open-cell-foundation-model/), [An open model bank for the rare tumours nobody has models for](https://onco.cc/ideas/idea-bio2-rare-tumour-model-bank/), [An open model of every cancer cell state, built from perturbation atlases](https://onco.cc/ideas/idea-moon-open-cancer-cell-state-model/), [An open organoid bank for cancers too rare to have models](https://onco.cc/ideas/idea-bio1-rare-cancer-organoid-bank/), [Audit animal studies for randomisation and blinding, published by institution](https://onco.cc/ideas/idea-tr2-arrive-audit/), [Automated combination discovery: patient-sample screens feeding Bayesian platform trials](https://onco.cc/ideas/idea-moon-automated-combination-discovery/), [Barcode patient-derived tumours to watch which clones win under each drug](https://onco.cc/ideas/idea-bio1-barcoded-avatars-clonal-fitness/), [Build a human model of the barrier that guards the brain fluid](https://onco.cc/ideas/idea-bio2-blood-csf-barrier-model/), [Build laboratory models of the organs cancer spreads to](https://onco.cc/ideas/idea-bio1-metastatic-niche-models/), [Every drug screen includes standard reference compounds whose performance is published](https://onco.cc/ideas/idea-tr2-reference-compound-panels/), [Every resistance mechanism found in a patient must be rebuilt in the laboratory](https://onco.cc/ideas/idea-bio1-reverse-translation-resistance-models/), [Grow blood-borne tumour cells to test drugs on the cells that actually spread](https://onco.cc/ideas/idea-bio2-ctc-culture-functional-testing/), [Grow each trial patient's tumour as organoids to decide which platform arm opens next](https://onco.cc/ideas/idea-tr2-organoid-coclinical-arms/), [Grow models from tumour cells in the blood when a biopsy is impossible](https://onco.cc/ideas/idea-bio1-ctc-derived-explants/), [Grow tumour organoids together with the patient's own immune cells](https://onco.cc/ideas/idea-bio1-organoid-immune-coculture/), [Hold organoid drug tests to the same standard as a diagnostic test](https://onco.cc/ideas/idea-bio1-organoid-assay-clinical-validation/), [Humanised mice with an immune system matched to the tumour donor](https://onco.cc/ideas/idea-bio1-immune-matched-humanised-mice/), [Implant a tiny device that tests twenty drugs inside the patient's own tumour](https://onco.cc/ideas/idea-bio2-implantable-microdevice-screen/), [In silico trials to choose the dose before the first patient](https://onco.cc/ideas/idea-bio1-in-silico-trials-dose/), [Keep a freshly removed tumour alive on a pump and test drugs in it](https://onco.cc/ideas/idea-bio1-ex-vivo-perfused-tumour/), [Linked human organ chips to predict side effects before people are dosed](https://onco.cc/ideas/idea-bio1-multi-organ-chip-tox/), [Make bespoke mouse cancer models in weeks with in vivo gene editing](https://onco.cc/ideas/idea-bio1-somatic-crispr-gemms/), [Make in vivo metastasis screens a required step in drug discovery](https://onco.cc/ideas/idea-bio2-metastasis-screen-standard/), [Multi-centre randomised animal trials before committing to a human trial](https://onco.cc/ideas/idea-bio1-multicentre-mouse-trials/), [Multi-laboratory preclinical trials as the standard for go/no-go decisions](https://onco.cc/ideas/idea-tr2-multilab-preclinical/), [Patient-derived organoids to pick ADC payloads](https://onco.cc/ideas/idea-organoid-guided-adc/), [Pet dogs with spontaneous cancer as a bridge before human trials](https://onco.cc/ideas/idea-bio1-comparative-oncology-dogs/), [Pick the laboratory model that matches the patient, not the one to hand](https://onco.cc/ideas/idea-bio1-model-patient-matching/), [Pre-register animal efficacy studies like clinical trials](https://onco.cc/ideas/idea-bio1-preclinical-preregistration/), [Pre-specified sample sizes for animal studies; no more 'representative' experiments](https://onco.cc/ideas/idea-tr2-animal-power-mandate/), [Prove the cell line is what you say it is, or the paper does not run](https://onco.cc/ideas/idea-bio1-model-authentication-mandate/), [Prove your cell lines are what you say they are, or the paper is not published](https://onco.cc/ideas/idea-tr2-cell-line-authentication-mandate/), [Reprogramme the stroma rather than remove it: second-generation stromal trials with a stromal biomarker and a survival endpoint](https://onco.cc/ideas/idea-pdac-stromal-reprogramming-not-depletion/), [Run the mouse or organoid trial at the same time as the human trial](https://onco.cc/ideas/idea-bio1-co-clinical-avatar-trials/), [Score every model system on how well it predicted real trial results](https://onco.cc/ideas/idea-bio1-model-predictivity-benchmark/), [Score every preclinical model by how often it predicted the clinical result](https://onco.cc/ideas/idea-tr2-model-report-cards/), [Self-driving laboratories that run the cancer biology hypothesis loop autonomously](https://onco.cc/ideas/idea-moon-self-driving-cancer-labs/), [Shared reference organoid and PDX panels that every lab can test against](https://onco.cc/ideas/idea-tr2-reference-model-panels/), [Test cancer drugs in old and unhealthy animals, not just young fit ones](https://onco.cc/ideas/idea-bio1-aged-comorbid-models/), [Test drugs on freshly cut slices of the patient's own tumour](https://onco.cc/ideas/idea-bio1-tumour-slice-cultures/), [Test drugs on the patient's own cancer cells when there is no trial to join](https://onco.cc/ideas/idea-bio2-functional-precision-rare/), [Tumour-on-a-chip with blood flow to test whether big drugs actually get in](https://onco.cc/ideas/idea-bio1-tumour-on-chip-penetration/), [Two-week pre-operative windows to compare combination biology head to head](https://onco.cc/ideas/idea-tr2-window-of-opportunity-triplets/), [Use patient organoids to check a cell therapy will work before infusing it](https://onco.cc/ideas/idea-bio1-organoid-cell-therapy-potency/), [Whole-patient digital twins validated in prospective randomised trials](https://onco.cc/ideas/idea-moon-validated-digital-twins/), [Zebrafish avatars for a drug answer within a week](https://onco.cc/ideas/idea-bio1-zebrafish-avatars/)
- cancers: [Colorectal cancer](https://onco.cc/cancers/colorectal/), [Glioma & glioblastoma](https://onco.cc/cancers/glioblastoma/), [Pancreatic ductal adenocarcinoma](https://onco.cc/cancers/pancreatic/)
- fronts: [Drug Discovery Platforms](https://onco.cc/fronts/drug-discovery/)
- technologies: [BH3 profiling (functional apoptosis testing)](https://onco.cc/technologies/bh3-profiling/), [CRISPR functional genomics](https://onco.cc/technologies/crispr-screens/), [Functional (ex vivo) drug testing](https://onco.cc/technologies/functional-drug-testing/), [Patient-derived organoids](https://onco.cc/technologies/organoids/), [Patient-derived xenografts](https://onco.cc/technologies/pdx-models/), [PDAC organoid pharmacotyping](https://onco.cc/technologies/pdac-organoid-pharmacotyping/)
- companies: [Champions Oncology](https://onco.cc/companies/champions-oncology/), [Curesponse](https://onco.cc/companies/curesponse/), [Recursion Pharmaceuticals](https://onco.cc/companies/recursion/), [SEngine Precision Medicine](https://onco.cc/companies/sengine/), [Xilis](https://onco.cc/companies/xilis/)
- institutions: [Atrium Health Wake Forest Baptist Comprehensive Cancer Center](https://onco.cc/institutions/wake-forest-cancer/), [Barbara Ann Karmanos Cancer Institute](https://onco.cc/institutions/karmanos/), [Broad Institute of MIT and Harvard](https://onco.cc/institutions/broad-institute/), [Cancer Center at Illinois](https://onco.cc/institutions/cancer-center-at-illinois/), [Chan Zuckerberg Biohub](https://onco.cc/institutions/cz-biohub/), [Cold Spring Harbor Laboratory](https://onco.cc/institutions/cold-spring-harbor/), [David H. Koch Institute for Integrative Cancer Research at MIT](https://onco.cc/institutions/mit-koch/), [Istituto di Candiolo IRCCS (FPO)](https://onco.cc/institutions/candiolo/), [National Cancer Institute (NIH)](https://onco.cc/institutions/nci/), [Rosalind and Morris Goodman Cancer Institute, McGill University](https://onco.cc/institutions/mcgill-goodman/), [The Francis Crick Institute](https://onco.cc/institutions/francis-crick/), [The Jackson Laboratory Cancer Center](https://onco.cc/institutions/jackson-laboratory/), [The Wistar Institute](https://onco.cc/institutions/wistar/), [UC Davis Comprehensive Cancer Center](https://onco.cc/institutions/uc-davis-cancer/), [UMC Utrecht Cancer Center](https://onco.cc/institutions/umc-utrecht/)
- key papers: [Comprehensive genomic profiles of small cell lung cancer](https://onco.cc/key-papers/paper-george-sclc-genomic-profiles-nature-2015/), [Defining a Cancer Dependency Map: which genes each cancer cell line cannot live without](https://onco.cc/key-papers/paper-depmap-tsherniak-cell-2017/), [Drug development: Raise standards for preclinical cancer research](https://onco.cc/key-papers/paper-begley-nature/), [Dual-targeted therapy with trastuzumab and lapatinib in treatment-refractory, KRAS codon 12/13 wild-type, HER2-positive metastatic colorectal cancer (HERACLES)](https://onco.cc/key-papers/paper-sartore-bianchi-heracles-trastuzumab-lapatinib-lancet-oncol-2016/), [Estimation of clinical trial success rates and related parameters](https://onco.cc/key-papers/paper-wong-biostatistics/), [Genotypic and histological evolution of lung cancers acquiring resistance to EGFR inhibitors](https://onco.cc/key-papers/paper-sequist-genotypic-histological-evolution-egfr-resistance-sci-transl-med-2011/), [Molecular determinants of resistance to antiandrogen therapy](https://onco.cc/key-papers/paper-chen-androgen-receptor-overexpression-antiandrogen-resistance-nat-med-2004/), [Reproducibility Project: Cancer Biology found that landmark preclinical results mostly shrank or vanished on replication](https://onco.cc/key-papers/paper-reproducibility-project-cancer-biology-elife-2021/)
- roadmaps: [Drug discovery roadmap: screening in mice → maps of dependency → designing in silico](https://onco.cc/roadmaps/drug-discovery-roadmap/)

---
JSON: https://onco.cc/api/v1/entities/b-preclinical-models.json