An App for Sore AIs: Research App Seeks to Predict Seasonal Allergies
Being in nature should be calming, amid chirping birds, buzzing bees, burbling brooks. But pollen, in a sneak attack, makes allergy sufferers erupt into a sneezing, watery-eyed mess.
It’s not easy to predict when seasonal allergies will strike — but it’s a question researchers at Palo Alto-based startup doc.ai are tackling with their first data trial.
Doc.ai built a mobile app that connects healthcare companies and medical researchers with smartphone users. App users elect to become research participants in data trials, sharing their anonymized information in exchange for health insights and compensation through blockchain tokens.
The first of these trials focuses on allergies, while upcoming trials span a range of medical research areas and diseases.
“We’re trying to accelerate research by enabling better and more predictive models when it comes to health and medical data,” said Sam De Brouwer, co-founder and chief operating officer at doc.ai.
The startup, a member of the NVIDIA Inception virtual accelerator program, is working with members of the healthcare industry like Anthem as well as research advisors from Harvard and Stanford Medical School.
Along with doc.ai researchers led by the company’s Chief Science Officer Jeremy Howard, these partners will be able to temporarily access anonymous user data to build AI-based predictive models.
Data Trials from the Comfort of One’s Own Phone
Doc.ai gives its app users the opportunity to participate in and benefit from medical research while retaining control of the data they choose to share. The company has built a Python-based data science platform for training and building models, and is using blockchain technology to safeguard the privacy and security of users’ medical data.
“People want to understand what’s happening with their data and be in full control,” De Brouwer said.
Its first data trial, run internally by doc.ai researchers, is building a deep learning network to predict seasonal allergies. The one-year trial is designed for 2,000 participants.
To create the predictive model for allergies, the researchers collect three types of data. First, users take a selfie through the mobile app. An AI model called Selfie To BMI, trained on 2 million faces, uses these selfies to predict the body-mass index from each user’s age, height, weight and gender.
Next, they enter their present and past zip codes. This location data is linked to 40 years of open source air quality and pollen information, which helps the model determine the kinds of allergens the trial participant is likely inhaling.
Lastly, the app connects to the user’s Fitbit or Apple HealthKit data to collect physical activity information. Participants self-report through the app when they experience allergy symptoms.
“We’re reducing the barrier to putting together research to a few weeks instead of a few months,” said De Brouwer. “Everything’s on the phone.”
Using NVIDIA Tesla V100 GPUs on the Google Cloud Platform, doc.ai researchers train their neural networks on the anonymized data that trial participants agree to share.
In the future, this de-identified data will be accessible to vetted data scientists outside of doc.ai. Participants must give informed consent about which aspects of their data are being shared for any data trials they sign up for.
Other research projects in the works span research areas like pain management, epilepsy, Crohn’s disease and Lyme disease. These future trials will incorporate new data points, too, including microbiome, genetic and blood test information.