By Paul Foster
It started with a pair of sunglasses.
In the spring of 2012, Brett Beaulieu-Jones was preparing to enter the doctoral program in genomics and computational biology at the Perelman School of Medicine when a friend, Jeffrey Impey, asked if he was up for a project. Impey had seen a pair of sunglasses in a movie and, wanting to get himself a pair, did what everyone does and did some online digging. Surprisingly, he found nothing. These days, though, when you can’t find something on the internet, generally what you find instead is an opportunity.
The quest for those sunglasses led the pair, with a third friend, Alex Loverde, to foray into the logistics of movie and television production, where Beaulieu-Jones applied the computational skills that are now helping him glean fresh insights from health data. In his doctoral studies, he is training computers to analyze huge samples of patient data. Through machine learning—helping the computers get better at the analysis on their own—Beaulieu-Jones hopes his team can more easily find disease subtypes in patients just by looking at the data in electronic medical records.
At the opposite end of the technology spectrum are the methods still in common use in Hollywood just a few years ago. While researching the movie and television industry, Beaulieu-Jones and his friends learned that production designers were still using binders to keep track of everything on set. They worked for the next few years to build Synconset, a company that developed software aimed at improving collaboration among production teams, an especially important feature as productions often include hundreds of people working in different locations—not great for a binder.
The Synconset team also developed an app that could receive a screenplay and perform a breakdown, turning the script into metadata. The metadata is used by crews to determine how many costumes will be needed, which characters should be wearing what costumes for which shots to maintain continuity, the consistency of design between shots for a scene, and so on. A hit out of the gates, the software was expanded to include props, makeup, hair, locations, and set decorations.
Jeffrey Impey, Alex Loverde, and Brett Bealieu-Jones
Beaulieu-Jones said the company grew organically, spread by word of mouth, and now, after a relatively short period of time, Synconset is used by half of all television and film productions. The effect they’ve had in Hollywood hasn’t gone unnoticed, either. In October 2016, the Synconset team flew to Los Angeles to accept a Primetime Engineering Emmy Award.
“I think we were the youngest group to win an Engineering Emmy,” Beaulieu-Jones said. “We were the only winners there who all brought their parents.”
Meanwhile, back at Penn, Beaulieu-Jones isn’t as involved in Synconset as he used to be. Instead, he’s now using his technical know-how to help make the most of the digital mountains of data we collect each day.
“Synconset has helped me learn how to build and manage large databases in complicated workflows,” he said. “Electronic medical records struck me as an area where there’s a huge amount of data, but due to the realities of the clinic, the data is complicated and noisy and we aren’t fully utilizing it for research yet.”
Beaulieu-Jones and team are studying how machine learning can potentially help diagnose patients with metabolic syndrome, a cluster of symptoms that places them at risk of diabetes or cardiovascular disorders, through the health data collected while they are treated at Penn. Patients come in for a variety of reasons, and doctors are focused on treating the priorities. So Beaulieu-Jones wants to develop the system to automatically analyze the many points of data collected and predict future issues a patient could face, ultimately allowing a physician to intervene earlier.
Digesting the amount of data created at Penn and accurately predicting future diagnoses isn’t possible for humans to do on their own, according Beaulieu-Jones, but thanks to enormous advancements in technology and a clever team from a broad, and unexpected, range of backgrounds, it’s becoming a reality.
This article was originally published on the Penn Medicine News Blog. Read a series of posts published in January about big data at Penn Medicine.