Using AI to identify genetic variants in tumors with DeepSomatic

Google Research · 9 months ago

Google Research and partners developed DeepSomatic, a machine learning tool using convolutional neural networks to identify genetic variants in tumor cells more accurately than existing methods across all major sequencing platforms. The tool achieved 90% accuracy on insertion-deletion variants using Illumina sequencing compared to 80% for the next-best method, and was trained on a new dataset called CASTLE created from six cancer cell lines sequenced with three different platforms. DeepSomatic can now be used by researchers and clinicians to identify cancer-driving mutations and inform precision medicine treatment plans.