Speakers at Biomedical Informatics Symposium Focus on Using Data Science To Improve Health
Posted in GUMC Stories
NOV. 18, 2015 – At the fourth annual Biomedical Informatics Symposium, speakers discussed some of the obstacles and opportunities for researchers striving to maximize the potential of biomedical informatics.
More than 350 attendees from academia, government and the private sector participated in the symposium on Oct. 16 at the Georgetown University Conference Center. In his opening remarks, Robert Clarke, PhD, dean of research at Georgetown University Medical Center (GUMC), described Georgetown as the home of bioinformatics.
In the 1950s, Margaret Dayhoff, a GUMC professor who would later be called “the mother of bioinformatics,” developed a single-letter code for amino acids to decrease the number of punch cards needed to describe amino acid sequences in the era of punch card computing. Her work led to the 1965 publication of the Atlas of Protein Sequence and Structure.
“She invented this single letter amino acid code which is used in probably every biochemistry textbook in every language on planet earth,” Clarke said. “That was the start of using computers to store, annotate, analyze and interpret biological information, in the context for her of protein sequence analysis and beginning to understand the evolution of individual proteins.”
ICBI Continues Tradition of Bioinformatics
Today, ICBI is continuing Dayhoff’s legacy at Georgetown. In addition to conducting bioinformatics and health IT research, ICBI provides research support services for the clinical-translational research community at Georgetown and beyond, said Subha Madhavan, PhD, ICBI director and associate professor of oncology at GUMC. ICBI also offers informatics education and training for early-career scientists and interested physicians, as well as high school, college and graduate students.
While working with multiple departments, faculty and staff at ICBI have received training in biomathematics, computational biology, computer science and biostatistics. “By design, the faculty members are from different departments but they come together in a multidisciplinary center to do their work and to inspire others to adopt their methods and techniques and technologies,” Madhavan said. “It’s so much fun to solve problems that way, when we bring multidisciplinary people together to focus on one problem.”
Madhavan identified several critical needs for researchers working in bioinformatics, such as the ability to use electronic health records for research purposes. As the recipient of a Clinical and Translational Science Award (CTSA), GUMC researchers have access to 3.4 million patient records, a number that will increase by partnering with larger national networks to drive clinical research. Other critical needs include focusing more attention on precision medicine and developing advanced data mining techniques.
Using Data to Determine Health Risks and Movie Preferences
Those working in biomedical informatics can draw inspiration from the way Netflix has worked to predict what movies its users will enjoy, said Isaac (Zak) Kohane, MD, PhD, chair of the department of biomedical informatics at Harvard University, during his keynote speech. In 2006, the company launched a contest to find an algorithm that would predict user ratings for films by publicly releasing data collected from users indicating what movies they watched and how they rated the movies.
“They are taking your whole history of movie watching and they’re taking the whole history of everyone else who watches movies and they’re clustering them into groups of people who have movie histories like you. And then they predict, on the basis of that, what’s your next move,” Kohane said. “They made all the data available for a contest for other data nerds to come up with the best predictor.”
“We should aspire to be as good in medicine as Netflix,” he added.
One example of the ways in which medical bioinformatics can be used to make predictions was addressed in a 2009 paper published in the British Medical Journal. The paper, which was co-authored by Kohane, described a computer model that could predict whether patients were facing domestic violence up to two years before doctors could make the diagnosis. Analyzing records collected from different emergency rooms, the model assigned patient visits with a value indicating how strongly the patient’s diagnoses correlated with abuse.
“It’s only when you start looking at individuals from this multi-axis perspective that you really start getting a good specific sense of who we are and that’s where the precision comes in,” Kohane said. “Because ultimately, we’re going to ask, who are you like so that we know what your diagnosis is, what your future path is and what are the right therapies.”