Nvidia, Scripps Research partner to advance AI for disease prediction

Nvidia and the Scripps Research Translational Institute announced Tuesday that they're partnering to advance the use of artificial intelligence for early disease prediction and prevention. More specifically, they'll establish a center of excellence to accelerate the creation of AI applications that use genomic and digital health sensor data.

So far, AI applications in medicine have largely focused on medical imaging, Kimberly Powell, vice president of healthcare at Nvidia, told reporters last week. While medical imaging is a powerful diagnostic tool, she said that AI needs to be applied to the medical data being collected from a growing number of sources. Data from DNA profiles or wearable technology, for instance, can go beyond diagnostics to help medical practitioners and researchers "think about the prevention of disease or the prediction of risk of disease in the first place."

Researchers from Nvidia and Scripps will work together to develop deep learning approaches to help improve mutation detection using genomics data. From mutation detections, deep learning can predict phenotypic information, which can be paired with digital sensor data to create new opportunities for disease prevention and intervention. Patients and clinicians are collecting more and more sensor data from devices like smartwatches, blood pressure cuffs and glucose monitors.

In addition to building datasets, the researchers will need to develop tools such as AI training frameworks that are friendly to genomics and sensor data, Powell said. Existing frameworks were built for text, speech or images, she explained, while "genomics and sensor data is going to look wildly different."

To prepare genomics data for a typical framework, it can be transformed into an image, Powell said. "That translation of sensor data into an image that a deep learning framewok can accept... that's something we can provide in a software tool," she said.

As researchers from Nvidia and Scripps build these sorts of tools, they'll make them publicly available for the research community at large.