Enhancing the value of genomic data

Genetic heritability is responsible for 30% of individual health outcomes, but is hardly used to guide disease prevention and care. Each individual carries 4-5 million genetic variants, each with varying influence on traits related to our health. The cost to sequence a genome has reduced drastically in recent years, and sequence data shows potential for ubiquitous use. However, the ability to read the sequence accurately and to meaningfully interpret it remain obstacles to broad adoption.

Improving the accuracy of genomic analysis


Sequencing genomes enables us to identify variants in a person’s DNA that indicate genetic disorders such as an elevated risk for breast cancer. DeepVariant is an open-source variant caller that uses a deep neural network to call genetic variants from next-generation DNA sequencing data. Learn more


Highly accurate genomes with deep neural networks


Despite rapid advances in sequencing technologies, accurately calling genetic variants present in an individual genome from billions of short, errorful sequence reads remains challenging. As published in Nature Biotechnology, DeepVariant significantly improves the accuracy in identifying variant locations, reducing the error rate by more than 50%.

DNA map

Winner in PrecisionFDA V2 Truth Challenge


DeepVariant won awards for Best Overall accuracy in 3 of 4 instrument categories in the PrecisionFDA V2 Truth Challenge. Compared to previous state-of-the-art models, DeepVariant v1.0 significantly reduces the errors for widely-used sequencing data types, including Illumina and Pacific Biosciences. Read the article


Identifying disease-causing variants in cancer patients


Researchers wanted to understand if incorporating automated deep learning technology would improve the detection of disease-causing variants in patients with cancer. In a cross-sectional study published in JAMA of 2,367 prostate cancer and melanoma patients in the US and Europe, DeepVariant found disease-causing variants in 14% more individuals than prior state-of-the-art methods.