Researchers at Dana-Farber Cancer Institute and MIT built OncoNPC, a model trained on the genetic mutations of 36,445 tumors with known origins, to predict the birthplace of cancers whose primary site is unknown.
On held-out tumors of known origin, its confident predictions were highly accurate. Applied to 971 cancers of unknown primary, it made a confident call for 41 percent of them, and it more than doubled the number of patients who could have received genetically guided treatment.
In a retrospective comparison, patients whose treatment matched OncoNPC's predicted origin tended to survive longer than those whose treatment did not, though this was an observational analysis, not a randomized trial, so it cannot prove the matched treatment caused the difference. The tool is not yet part of routine cancer care.
Read the original study
Moon et al., Nature Medicine, 2023 · doi.org/10.1038/s41591-023-02482-6
Software read the mutations and named the cancer's hometown.
Source
- Moon, I., LoPiccolo, J., Baca, S. C., Sholl, L. M., Kehl, K. L., Hassett, M. J., Liu, D., Schrag, D., & Gusev, A. (2023). Machine learning for genetics-based classification and treatment response prediction in cancer of unknown primary. Nature Medicine, 29(8), 2057–2067. https://doi.org/10.1038/s41591-023-02482-6