Human Cancer Models Initiative Builds 665 Cancer Models With 97.8% Genetic Fidelity
Updated
Updated · Nature.com · Aug 5
Human Cancer Models Initiative Builds 665 Cancer Models With 97.8% Genetic Fidelity
3 articles · Updated · Nature.com · Aug 5
Summary
665 patient-derived cancer models from 637 patients across 25 malignancies were generated by the Human Cancer Models Initiative and released with molecular profiles, clinical data and software tools for broad research use.
421 matched tumour-model comparisons showed 97.8% genetic concordance and 95% epigenetic concordance, while RNA analyses found 92% transcriptional agreement despite at least a year of culture.
153 models cover rare cancers, 71 come from primarily non-European ancestry donors, and 522 include extensive clinical annotations, addressing long-standing gaps in diversity and patient context.
Single-cell and media analyses showed most models preserved tumour cell states, but some drifted under culture conditions—especially glioblastoma grown in certain media—highlighting limits and ways to improve fidelity.
Compared with the CCLE, HCMI expanded public coverage of colorectal, pancreatic and oesophageal cancers and, when combined with CCLE, provided at least one high-fidelity model for 77.3% of TCGA tumours.
Why did it take a decade to create cancer models reflecting true human diversity, and what hidden cures might they now reveal?
If these new 3D cancer models perfectly mimic human tumors, could they finally end the need for animal testing in drug development?
The Human Cancer Models Initiative in 2026: 665 Next-Generation Models Transforming Global Oncology Research and Precision Medicine
Overview
Traditional 2D cancer cell cultures and animal studies have long failed to accurately mimic the complexity of real human tumors, leading to high drug failure rates and wasted resources. The Human Cancer Models Initiative (HCMI) addresses these problems by developing next-generation 3D models, such as organoids and neurospheres, which better preserve natural tissue architecture and stem cell populations. These models can quickly self-organize and closely match patient tumors, making them highly useful for predicting drug responses and studying resistance, as seen in glioblastoma research. However, challenges remain, including the lack of immune and stromal cells in standard organoids and slow turnaround times for clinical use.