Google AI Tool DeepSomatic Identifies Genetic Drivers of Cancer with Precision

 

Google AI tool pinpoints genetic drivers of cancer

Google introduces DeepSomatic, a breakthrough AI that identifies cancer-causing mutations in tumor DNA more accurately than ever before. Published in Nature Biotechnology, the tool could reshape precision medicine and cancer treatment.

The challenge of cancer genetics

Cancer begins when the genetic systems controlling cell division malfunction. To design effective, personalized therapies, doctors must identify the genetic mutations fueling tumor growth. Genome sequencing already helps target treatments, but distinguishing true mutations from sequencing noise remains a challenge.

Most cancers are caused by somatic variants—mutations acquired after birth, not inherited. These arise from DNA damage or random replication errors and can exist at low frequencies, making them difficult to detect.

How DeepSomatic works

DeepSomatic compares tumor DNA with normal DNA from the same patient to pinpoint mutations unique to the tumor. It transforms raw genetic data into image-like matrices analyzed by a convolutional neural network (CNN), distinguishing real mutations from errors.

DeepSomatic even works in “tumor-only” mode, making it invaluable for cancers like leukemia where healthy samples are unavailable.

Training with the CASTLE dataset

Google, the UC Santa Cruz Genomics Institute, and the National Cancer Institute created a benchmark dataset called CASTLE using breast and lung cancer samples sequenced across three platforms. This multi-platform dataset minimizes bias and enhances accuracy.

Outperforming existing tools

DeepSomatic surpassed leading methods in identifying complex mutations like insertions and deletions (Indels).

• Illumina data: 90% F1-score (vs. 80% next-best)
• PacBio data: 80%+ F1-score (vs. under 50%)

The AI also excelled on degraded or historical samples, such as FFPE-preserved tissues and exome-only sequencing, maintaining strong accuracy even in low-quality data.

Beyond its training scope

DeepSomatic successfully identified variants in cancers it wasn’t trained on. It located key mutations in glioblastoma and found ten new variants in pediatric leukemia samples from a collaboration with Children’s Mercy Kansas City.

The tool not only confirms known mutations but reveals new genetic clues, paving the way for future cancer therapies.

DeepSomatic represents a major leap in AI-powered genomics. By detecting cancer-driving mutations with unprecedented precision, it could accelerate personalized medicine and inspire new treatment discoveries.

Based on Google's publication in Nature Biotechnology and collaborations with UC Santa Cruz Genomics Institute and the National Cancer Institute.

Google AI DeepSomatic Cancer Research Genomics AI in Medicine Precision Medicine