Google announced DeepSomatic, an AI tool that can more accurately identify cancer-related mutations in a tumor’s genetic sequence.
Cancer occurs when the controls governing cell division malfunction. Finding the specific genetic mutations that promote tumor growth is essential to developing effective treatment plans. Doctors now routinely sequence the genomes of tumor cells from biopsies to inform treatments that target how specific cancers grow and spread.
The study, published in Nature Biotechnology, introduces a tool that uses convolutional neural networks to identify genetic mutations in tumor cells with greater accuracy than current methods. Google has made both DeepSomatic and the high-quality training datasets created for it openly available.
Challenges of somatic mutations
Cancer genetics are complex. While genome sequencing uncovers genetic mutations in cancer, it is difficult to distinguish between real mutations and sequencing errors, and AI tools are welcome if they can help. Most cancers are caused by “somatic” mutations acquired after birth, rather than “germline” mutations inherited from parents.
Somatic mutations occur when environmental factors, such as ultraviolet light, damage DNA, or when random errors occur during DNA replication. These mutants can alter normal cell behavior, leading to uncontrolled replication and promoting cancer development and progression.
Identifying somatic variants is more difficult than finding heritable ones because somatic variants can be present at low frequencies within tumor cells, sometimes at rates lower than the sequencing error rate itself.
How DeepSomatic works
In clinical practice, scientists sequence both tumor cells from biopsies and normal cells from patients. DeepSomatic finds differences and identifies mutations in tumor cells that are not inherited. These changes reveal what is driving tumor growth.
This model converts raw genetic sequence data from both tumor and normal samples into images representing different data points, such as sequence data and sequence data along chromosomes. Convolutional neural networks analyze these images, distinguishing between the standard reference genome, an individual’s normal genetic variations, and cancer-causing somatic mutations, and removing sequencing errors. The output is a list of cancer-related mutations.
DeepSomatic can also operate in “tumor-only” mode when a normal cell sample is not available. This often occurs with blood cancers such as leukemia. This makes this tool applicable to many research and clinical scenarios.
Training more accurate AI cancer research tools
Training accurate AI models requires high-quality data. As an AI tool, Google and its partners, the University of California Santa Cruz Genomics Institute and the National Cancer Institute, have created a benchmark dataset called CASTLE. They sequenced tumor cells and normal cells from four breast cancer samples and two lung cancer samples.
These samples were analyzed using three major sequencing platforms, and a single accurate reference dataset was created by combining the outputs and removing platform-specific errors. This data shows that cancers of the same type can have vastly different mutational signatures, providing information that can help predict a patient’s response to a particular treatment.
The DeepSomatic model outperformed other established methods on all three major sequencing platforms. The tool excelled at identifying complex mutations called insertions and deletions, or “indels.” For these variants, DeepSomatic achieved an F1 score of 90% on Illumina sequencing data. This was compared to 80% for the next best method. The improvement was even more dramatic in Pacific Biosciences’ data, with DeepSomatic scoring over 80%, while the next best tool scored less than 50%.
The AI performed well when analyzing difficult samples. The test included breast cancer samples stored in formalin-fixed paraffin-embedded (FFPE), a common method that can cause DNA damage and complicate analysis. It was also tested with data from whole exome sequencing (WES), a more affordable method that sequences only the 1% of the genome that codes for proteins. In both scenarios, DeepSomatic shows better performance than other tools, demonstrating the utility of DeepSomatic for analyzing low-quality or historical samples.
AI tools for every cancer
The AI tool showed that it can apply its learning to new types of cancers for which it has not been trained. When used to analyze samples of glioblastoma, a highly malignant brain tumor, they were able to pinpoint several mutations known to cause the disease. In partnership with Children’s Mercy in Kansas City, researchers analyzed eight childhood leukemia samples and discovered previously known variants, as well as identifying 10 new variants, even though they used tumor-only samples.
Google hopes that research institutions and clinicians will adopt this tool to better understand individual tumors. Detecting known cancer variants can help guide the selection of existing treatments. Identifying new ones may lead to new treatments. The goal is to advance precision medicine and provide more effective treatments to patients.
SEE ALSO: MHRA quickly adapts to next wave of AI tools for patient care
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events such as Cyber Security Expo. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
