DeepSomatic accurately identifies genetic variants in cancer

Cancer is a genetic disease, and identifying the mutations in cancer cells is critical for delivering more precise treatments and better outcomes for patients. That’s why we’re proud to introduce DeepSomatic, a new machine learning model developed with our partners, including UC Santa Cruz and Children’s Mercy, that accurately identifies genetic variants in cancer cells.

DeepSomatic uses convolutional neural networks to differentiate between inherited variants and the acquired somatic variants that drive cancer, even working on complex samples like pediatric leukemia and glioblastoma. The tool is proven to outperform current methods at spotting these key variants, particularly insertions and deletions. We have made both the tool and its high-quality training dataset openly available to speed up cancer research globally.

Google Research is making foundational breakthroughs that have real, tangible impact for people. That’s why we do this work, because the path to the future is founded in research that can make reality better for people. We look forward to seeing how the research community uses it to accelerate cancer research, for everyone.