Simon Eng Wins Donnelly Thesis Prize for Machine Learning Research that Could Improve Lives of Children with Arthritis
At a recent conference on childhood arthritis, Simon Eng met not only other researchers in the field but also patients who shared what it’s like to live with aching joints and their hopes for the future.
“These kids are very passionate about finding appropriate treatment,” says Eng, who during his PhD in the Donnelly Centre developed a computational tool for classifying patients into distinct disease groups to help doctors better tailor treatment that has fewer side effects. “It was amazing to see the excitement in their eyes about how our work could help them.”
Childhood arthritis has no cure and treatment consists of pain medication and immune-suppressing drugs which do not work in everyone and may have side effects.
For his research, Eng won the 2019 Donnelly Thesis Prize, awarded annually for the best doctoral research completed at the Centre. He was co-supervised by artificial intelligence researcher Quaid Morris, of the Donnelly Centre, and Dr. Rae Yeung, rheumatologist and Senior Scientist at the Hospital for Sick Children.
“On behalf of the award committee, I would like to congratulate Simon on this deserving award,” says Jason Moffat, Chair of the award committee with Donnelly Centre investigators Cindi Morshead, William Ryu and Aaron Wheeler as members.
“This was a competitive year with amazing candidates but Simon’s work provides the basis for a new way of understanding and potentially treating childhood arthritis,” says Moffat.
To Eng, the award is “a celebration of a fruitful collaboration” that saw immunologists, clinicians, and AI experts join forces to tackle a major problem in childhood health.
“I am very grateful to my supervisors Quaid and Rae who are leading researchers in their fields,” he says. “With co-supervision you learn there are multiple ways of approaching the same problem and you get to choose the best of what both of them bring to the table.”
One challenge in straddling the worlds as disparate as immunology and computer science is finding a common language. But Eng was up to the task having earned degrees in both as an undergraduate at the University of British Columbia in Vancouver. “It meant that I was able to both do the data analysis and interpretation and facilitate discussion between the two teams.”
"With co-supervision you learn there are multiple ways of approaching the same problem and you get to choose the best of what both of them bring to the table" - Simon Eng, winner of the Donnelly Thesis Prize.
Childhood arthritis occurs when the body mistakes its own cells for foreign invaders and attacks joint tissue, causing swelling and throbbing pain. With about 24,000 Canadian children suffering from it, it is the most common childhood rheumatic disorder.
Furthermore, children can experience very different symptoms but so far there was no good way of classifying them into distinct subcategories to reflect the underlying biology of the disease. As a result, patients can end up receiving medication that is not suitable for them.
“The reason we are so concerned about patient classification is from the treatment perspective,” says Eng. “If we can come up with a classification that encapsulate the biology of the disease, this could provide hints as to how we can treat these different patient groups to give them most specific treatment as possible to avoid side effects.”
Eng came up with a computational approach based on unsupervised machine learning, a type of artificial intelligence, that can classify patients into seven distinct categories based on the pattern of inflamed joints. The algorithm can also predict which patients will go on to outgrow their arthritis and can be spared the more aggressive medications, as described in a landmark study published earlier this year in the journal Plos Medicine.
Collaborating with several Canadian consortia, Eng had access to a wealth of clinical, demographic, and biological patient data, including gene and protein expression. By applying machine learning to these diverse datasets, he was able to identify patterns which distinguish distinct disease subtypes. Eng remained at SickKids to combine all his findings into a single disease classification system. His hope is to develop an app to help doctors diagnose patients on the spot after initial examination.
“One of the things we are working on is to build an app so that a clinician can examine a kid in the clinic, enter all the joints that have arthritis and ideally the system will spit out what the joint pattern is maybe even suggest a course of treatment.”
His next career move? “We’ll see what happens,” he says, revealing only that a career in a health tech is on the table.