NEW YORK – Duality Technologies said Monday that it has been awarded a $1 million contract from the US Defense Advanced Research Projects Agency (DARPA) to develop machine-learning technology for researching genomic susceptibility to COVID-19.
A key feature of the technology is that it must preserve individual privacy while using encrypted data from multiple sources, the Newark, New Jersey-based company said. To this end, Duality will apply homomorphic encryption, a method of encoding data as ciphertext to allow computation without decryption, to the data.
The firm said that its ability to train machine-learning algorithms on encrypted data will help researchers pool genomic and clinical data from multiple institutions to build models that can predict patients' susceptibility to COVID-19.
The DARPA contract also will support research into the sharing of intelligence on cyber threats, again using homomorphic encryption.
"This contract award supports Duality's commitment to promoting responsible, privacy-preserving use of big data for the public good and for national security," Duality Cofounder and CTO Kurt Rohloff said in a statement. "Secure machine-learning model training is the next frontier in privacy-preserving analytics, and we are excited to make this advanced capability applicable to real-world challenges."
The company demonstrated the efficacy of its homomorphic encryption in a paper published in May in the Proceedings of the National Academy of Science. For that study, Duality and researchers at Dana-Farber Cancer Institute ran a GWAS on a dataset containing genotypic and phenotypic information on more than 25,000 patients and produced results 30 times faster than a previous method, without sacrificing accuracy.