The hackathon winning team (from left) Ben Hsu, BS; Nandita Damaraju, MS; Jo Varshney, DVM, PhD; and Teng Gao, BS. Photo courtesy of Jo Varshney
A UC San Francisco cancer researcher has led a team of data scientists and engineers to win a first-of-its-kind Artificial Intelligence (AI) Genomics Hackathon competition.
Jyotika ‘Jo’ Varshney, DVM, PhD, a UCSF postdoctoral fellow, and her three teammates were named the winners at the event, which challenged participants to analyze a real patient’s genomic data using AI and other computational methods in order to advance the understanding of a rare genetic disease called neurofibromatosis type 2, or NF2.
“There is much to be gained from bringing AI experts, data scientists and biologists together, and this interdisciplinary approach was one of the best aspects of the hackathon,” Varshney said. “There are so many new tools available, and to find meaningful discoveries, we absolutely cannot do cancer research the way we did 10 years ago.”
Entrepreneur Recognized Need for AI Approach
Recognizing the need for new approaches utilizing the latest AI technologies, Silicon Valley entrepreneur and NF2 patient Onno Faber released his genomic data for event participants to access and build models for identifying new therapeutic targets for his disease, which affects one in 40,000 people worldwide.
NF2 is caused by mutations in the NF2 gene that lead to the faulty production of a certain protein that can no longer regulate cell growth and division. As a result, Schwann cells, which surround and insulate neurons in the brain and spinal cord, multiply too frequently and form tumors in the nervous system.
Varshney, along with teammates Ben Hsu, BS, a software engineering intern at SigOpt, Nandita Damaraju, MS, a data scientist at Thermo Fisher, and Teng Gao, BS, a computer scientist, used an AI approach called deep learning to extract interaction networks from complex genomic data.
Bringing About Sea Change in Cancer Research
To get the most out of the smaller datasets available for rare diseases like NF2, they attempted a transfer learning approach and leveraged a larger dataset from a more common disease in order to make meaningful biological predictions for NF2 tumors. They were able to extract key biological principles from the neural network trained on Faber’s genomic data.
Read more at UCSF.edu