Korea University computer students rank among the top in the cancer precision medicine DREAM Challenge
Big Data and artificial intelligence combined to provide personalized drug treatments for cancer patients
Professor Jaewoo Kang’s research team develops a personalized anticancer drug combination algorithm
▲ Front row, from left: Minhwan Yu (M.A. course), Professor Jaewoo Kang, Sungjoon Park (M.A. course)
Back row, from left: Kwanghoon Choi (M.A. course), Heewon Lee (M.A. course), Minji Jeon (Ph.D. course), Hyeokyoon Chang (M.A. course), Sunkyu Kim (M.A. course)
Professor Jaewoo Kang’s research team from the Department of Computer Science and Engineering at the College of Informatics was ranked near the top in their field at the international biomedicine competition run by Dream Challenges. The results of their work put them in the top 3 percentile, allowing them to beat off the challenge of world class colleges such as Stanford.
‘DREAM Challenges’ is a non-profit organization that invites researchers from across the world to compete in various 'challenges' by proposing solutions to biomedicine-related problems, in effect constituting a scientific crowdsourcing effort. Within set criteria and using the datasets provided, global research teams participating in the various DREAM Challenges pit themselves against each other. At this competition, the multinational pharmaceutical company, AstraZeneca, and the world’s largest genome research institute, the Sange Institute, provided the datasets and posed three questions which invited participants to predict which combination of anticancer drugs would be most effective in treating cancer patients.
The cause of cancer differs from patient to patient, and because of this, anticancer treatment effectiveness likewise varies. Although patients receive personalized prescriptions based on their genome characteristics, resistance to single agent anticancer drugs causes many patients’ treatment effectiveness to decrease over time, often resulting in relapse. Precision medicine’s key task is to overcome resistance to anticancer drugs and to maximize treatment effectiveness by discovering the optimal mix, and finally to administer a sometimes complex cocktail of drugs. It is a daunting problem because it is not easy to analyze both the enormous amount of genome data that a single patient possesses and the possible effects of hundreds of anticancer drugs.
The algorithm developed by Professor Jaewoo Kang’s research team devises various anticancer drug combinations for different types of cancer patient, and based on this it recommends the optimal combination of anticancer drugs for specific cancer patients. During this process, ‘machine learning’ technology is applied, allowing the algorithm to learn for itself. This process is aided by the incorporation of information automatically extracted from hundreds of articles in the biomedicine literature, which enables the algorithm to become continuously ‘smarter’.
Professor Kang said, “Real issues in the medical field were resolved by combining the latest machine learning technology with Big Data”. He continued, “I hope that this research into how to predict the treatment effectiveness of drug combinations not only allows us to understand drug resistance and patient relapse, but leads to improved cancer treatment rates.”
Despite all members of Professor Kang’s research team (Minji Jeon, Sunkyu Kim, Sungjoon Park, Heewon Lee, Hyeokyoon Chang, Minhwan Yu) being Computer Department students, they competed in this DREAM Challenge against more than 70 world class research teams, and placed between 2nd and 4th in their responses to each of the three Dream Challenge questions. In this same competition Stanford placed 7th overall and the MIT team ranked 13th.
Professor Kang’s research team presented their research results at the RECOMB/ISCB Conference on RSG with DREAM Challenges in Phoenix, Arizona on November 9th. In addition, DREAM Challenge competition results are scheduled for publication in the most prestigious journal in this field, Nature Biotechnology.