Blue spiral representing NET project

SomaticNET

Neural-network Evaluation of Tumor Variants

Request access

Results

96% accuracy in variant calls from ground truth dataset. Early results published here.

Access

SomaticNET is shared freely on a non-commercial license. Request access from this page.

Challenge

Call somatic variants in cancer DNA at extreme accuracy.

Methodology

Create a mutation caller building on Google's DeepVariant, but with an entirely new architecture for cancer samples.

Interlacing Personal and Reference Genomes for Machine Learning Disease-Variant Detection

Variants in DNA underlie traits inherited from our parents, define the difference between two individuals of the same species and are the root cause of diseases such as cancer. Analysis of these variants enables the identification of potentially fatal diseases and conditions. Variant detection is the foundation of targeted cancer therapy and personalized medicine. If we get variant detection wrong, then the drugs we choose could be wrong, and patients may suffer. In this paper, CCG has developed a new class of variant detection algorithm, based on computer vision. In benchmark tests with ground truth datasets, our software already outperforms those developed by Google and the Broad Institute of Harvard and MIT. Our tool is available for free, for non-commercial use.

View the peer reviewed paper →