Search
  • Announcement
  • Artificial Intelligence

CIFAR researchers apply AI in the fight against COVID-19

by Krista Davidson May 12 / 20

CIFAR’s AI and COVID-19 Catalyst Grants Program in partnership with the Natural Sciences and Engineering Research Council of Canada (NSERC), the Ontario Government, Microsoft, and Genome Canada.

Launched on March 23, 2020, the program provides nearly $300,000 in total funding to innovative, high-risk, high-reward ideas and projects that address the current COVID-19 pandemic. The AI Catalyst grants are funded, in part, thanks to the generosity of the Max Bell Foundation and many individual donors. 

Here are example of some of the groundbreaking research projects selected as part of the CIFAR AI and COVID-19 Catalyst Grants Program.

line

Alberta team uses AI to find and protect vulnerable populations

The Canadian Government’s official guidance, which mirrors that of the guidance in the U.S. and elsewhere, is that certain groups—the elderly, those with pre-existing conditions, the immunocompromised—are at greater risk of developing severe COVID-19 symptoms. 

Data scientist Randy Goebel and Canada CIFAR AI Chair Martha White believe that, by analyzing large amounts of comprehensive health data that includes COVID-19 status, they will be able to tease out subtle connections and identify other vulnerable segments of the community.

“With secure access to testing data, we can answer questions from epidemiologists and public health officials, like “Is there any prevalence of positive testing in diabetics?” or “Is there a different pattern in indigenous communities?” says Goebel, who is leading an Artificial Intelligence (AI) project based in Edmonton that is training computer models to detect patterns in data from the Alberta Health Services to predict hot-spots and at-risk populations.

“What we want to do in Alberta is try to exploit the fact that we have one healthcare delivery system for 4.5 million people,” says Goebel, a member of CIFAR’s Pan-Canadian AI Strategy National Program Committee, a professor of computer science at the University of Alberta, and co-founder of Amii.

With the help of medical doctors, sociologists, and ethicists at the University of Alberta and Alberta Health Services, Goebel and White are working to gain simple, uniform, secure access to health data from across the province, apply machine learning models, and communicate their findings to the broader public. 

Their graduate students, supported by CIFAR catalyst funds, will formalize how the data will be gathered, analyzed, and disseminated.

“We want to help people understand where they fit into the current public health situation as individuals,” says Goebel. “What questions might they want to ask, and how can they participate in a community to help understand their role?”

Collaborators:
Daniel C. Baumgart (University of Alberta), Martha White (Canada CIFAR AI Chair, University of Alberta, Amii), Randy Goebel (Amii, University of Alberta), Geoffrey Rockwell (Kule Institute for Advanced Study, University of Alberta), Robert Hayward (Chief Medical Information Officer, Alberta Health Services), Shy Amlani, (Virtual Health), Jonathan Choy (Virtual Health), Sara Webster (Virtual Health), and Sarah Hall (Virtual Health)

line

AI research to assess the stress of COVID-19 on children and families

First-of-its-kind AI-enabled mobile app to help clinicians identify and treat stress.

While little is known about COVID-19, the latest studies from China suggest that children are less likely to experience severe symptoms of the disease, such as respiratory distress. However, the disease has had a profound impact on children and their families. Coping with social isolation and the demands of home-schooling has created new challenges. The role of stress in increasing a child’s susceptibility to contracting COVID-19 is not well understood.

Canada CIFAR AI Chairs Anna Goldenberg (CIFAR fellow, Child & Brain Development Program, Vector Institute, University of Toronto, Hospital for Sick Children) and Marzyeh Ghassemi (Vector Institute, University of Toronto), together with the team of pediatricians and tech experts are using machine learning to assess the broad implications of COVID-19 on children and their families. With support from the CIFAR AI and COVID-19 Catalyst Grants initiative, the study is the first of its kind to use artificial intelligence (AI)-enabled technology to generate a better understanding of the emotional and physical toll the disease takes on families.

"This project is a rare opportunity to examine how families acquire, experience, and hopefully recover from COVID-19," says Marzyeh Ghassemi, an assistant professor at the University of Toronto. "Learning about recovery in children is also important, because there may be communication barriers in younger children about levels of pain, or degree of fatigue."  

In collaboration with Evidation Health, a healthcare technology company at the forefront of developing secure platforms for wearable and mobile apps for medical purposes, the team is gearing up to deploy a mobile app designed to help clinicians understand the familial stress associated with COVID-19. The app leverages AI to assess the physiological stress of children and families, and measures health factors such as sleep, anxiety and energy levels, as well as physical activity.

The data can help health professionals determine stress levels in children and their families and enable them to allocate resources to help families cope with the stress associated with intervention methods.

“I’m thrilled to be part of this excellent team of clinical and technical experts,” says Goldenberg,  Varma Family Chair of Medical Bioinformatics and Artificial Intelligence and assistant professor at the University of Toronto. “This collaboration has been essential in developing a unique tool that can assess the health risks to children and their families during this critical time.”

The app draws on data provided by Canada’s largest children’s cohort study, TARGet Kids!, led by Jonathon Maguire and Catherine Birken, including rich information such as sociodemographics, age, health behaviors, nutrition and other information. The research will measure a variety of factors, including respiratory tract symptoms, preventative efforts such as social isolation and physical distancing, as well as data provided by smartphones. 

Collaborators:
Jonathon Maguire (Hospital for Sick Children), Anna Goldenberg (Canada CIFAR AI Chair and CIFAR Child and Brain Development program, Vector Institute, University of Toronto, Hospital for Sick Children), Marzyeh Ghassemi (Canada CIFAR AI Chair, Vector Institute, University of Toronto), Catherine Birken (Hospital for Sick Children), Peter Jüni (St. Michael’s Hospital) Kevin Thorpe (Sunnybrook Hospital), Charles Keown-Stoneman (St. Michael’s Hospital), Mary Aglipay (St. Michael’s Hospital)

line

Machine learning and ultrasound for COVID-19 pneumonia diagnosis

As the COVID-19 pandemic progresses, research shows that the disease can have a wide ranging severity—from a mild fever to pneumonia and death. Doctors are eager to find ways to detect severe cases as early as possible, as early identification and treatment leads to better outcomes

Machine learning may soon help physicians diagnose pneumonia with ultrasound scanners. Russell Greiner, a frequent CIFAR collaborator, professor in the Department of Computing Science at the University of Alberta, and fellow at Amii, plans to use machine learning to produce a model that will interpret ultrasound scans and help to diagnose pneumonia in COVID-19 patients. Greiner is part of a team which includes radiology experts at the University of Alberta and data scientists at Edmonton-based Medo.AI. They will build on decades of experience in medical informatics and previous successes diagnosing hip dysplasia and liver problems with AI and ultrasound.

Greiner says that the most reliable way to detect pneumonia is through computed tomography (CT) scans. However, the machines that perform these scans are immovable, cost hundreds of thousands of dollars, and require specially-trained personnel. Researchers also worry that the need to move patients to a central scanner is particularly risky in the case of an infectious disease like COVID-19.

The team is looking to use ultrasound scanners, a much smaller medical device, to detect pneumonia in people with COVID-19. “Ultrasounds are handheld, inexpensive, and relatively easy to use,” says Greiner. “The downside is that they produce a much less precise signal. It's much harder to interpret.”

“There's a chance I can actually make a difference in reducing mortality and morbidity,” says Greiner. “The whole world is affected by COVID-19. If there's a way we can play a small part, make a small contribution to reduce the suffering, to increase the diagnostic accuracy, to find ways to improve this horrible condition, I'm delighted to spend my time doing it.”

Collaborators:
Kumaradevan Punithakumar (University of Alberta), Russell Greiner (University of Alberta, Amii), Jacob Jaremko (University of Alberta), Nathaniel Meuser-Herr (Upstate Health Care Center, NY), Dornoosh Zonoobi (MEDO.ai)

line

Tracking mental health during the COVID-19 pandemic

Faced with hurdles such as home schooling, unemployment and social distancing from friends, colleagues and families, many people around the world are experiencing new and worsening mental health issues.

New reports from Twitter show that its daily active user base has grown by 23 per cent. It seems that many people are turning to social media to forge lost connections and engage in public discourse.

Alona Fyshe, a Canada CIFAR AI Chair and fellow of the CIFAR Learning in Machines & Brains program based at the University of Alberta, is working with a team of computer scientists to apply artificial intelligence (AI) techniques to social media to better understand these challenges and what impact they have on our mental health. 

The goal is to build a tool that can detect the drivers of mental health issues associated with the pandemic, and in the event of a second wave, prepare and anticipate techniques to cope with them.

Fyshe says the impetus for the project was in hearing that grocery stores were able to get a pulse on how people were feeling based on their buying habits.  Early on, consumers wanted to ensure basic necessities like toilet paper were met, then as people turned to baking, flour became an in-demand commodity.

“We thought, parallel to how people are behaving is what they are talking about on social media,” says Fyshe, a fellow at Amii and an assistant professor at the University of Alberta’s faculty of science.

Working with computer scientists Dan Lizotte at Western University and Rumi Chunara at New York University, the team is scraping Twitter for language about COVID-19 related to different mental health states to get a sense of the different topics that correlate with both negative and positive mental health issues. These include home-schooling, unemployment, frustration in accessing unemployment benefits, but also positive feelings associated with family time and baking.

“This will build on our work that looks at discourse within vulnerable populations on social media,” says Lizotte.

“There are ways to track the positive outcomes, like baking, and there are ways to track sadness and deprivation. Our project is using social media to look at how people are feeling and why, and help public health agencies  prepare for the concerns that come with mental health episodes,” says Fyshe.

The team is planning to share their findings through an online visual analytics and rapid reporting system that will identify the real-time topics that are front of mind for social media users.

Collaborators:
Alona Fyshe (Canada CIFAR AI Chair, CIFAR Learning in Machines & Brains program, Amii, University of Alberta), Daniel Lizotte (Western University), Rumi Chunara (New York University)

line

Using AI to find effective COVID-19 drugs

Researchers are using the latest machine learning techniques find existing drugs that may treat COVID-19

According to the Mayo Clinic, because SARS-CoV-2 is a new virus in humans, there is no existing immunity to it. With vaccines in development, but estimated to be at least a year away, many researchers are turning to pharmaceuticals—chemicals that can block the virus’s ability to infect and destroy lung tissue—to find effective treatments for COVID-19.

A recent study by the Tufts Center for the Study of Drug Development found that developing new drugs from scratch takes many years and billions of dollars. Instead, the scientific community is looking to the existing medicine cabinet of approved drugs, and using machine learning to predict which may be effective at treating COVID-19. 

Medical knowledge graphs

One approach, taken by a team out of Montreal led by Canada CIFAR AI Chairs Jian Tang, William Hamilton, and Yoshua Bengio, is to use a medical knowledge graph. This type of database, a version of which has been in development by members of the team for over a year, integrates a vast array of relationships between diseases, proteins, and drugs. The graph captures the known interactions between thousands of diseases, the proteins they affect, and the drugs that work on them. “We need to merge multiple different data sources to build a big, comprehensive knowledge graph, which involves multiple different entities and relationships,” says Tang.

The team will use machine learning to find drugs that affect the same proteins targeted by COVID-19, in a similar way to how social networks recommend friends. “In social networks, for example, you want to recommend friends based on existing links,” says Tang.  “In this case, it’s the same intuition. Based on the existing links between the drugs and proteins and the proteins and disease, we try to predict those new links.” 

Testing drug candidates in cells

Another approach, championed by a Toronto team led by the labs of CIFAR Azrieli Global Scholar Jean-Philippe Julien (University of Toronto, Sick Kids) and Costin Antonescu (Ryerson University) plans to test AI-identified drug candidates in cellular assays. With labs already set up to investigate the interactions between drugs, cells, and viruses, Julien and Antonescu are turning their attention towards SARS-CoV-2.

They are collaborating with two local drug development companies, Cyclica and Phoenox, who have already used AI to identify a list of candidate drugs that may be effective against COVID-19. In particular, they have found molecules that may interrupt the virus’s ability to gain entry into cells and others that may limit respiratory distress.

The catalyst funds will be used to purchase reagents and support a postdoctoral fellow. With tight collaboration between the partners, the group plans to work iteratively, feeding new information about protein interactions back to the AI team, who will predict more molecules worth testing. Within four months, the team hopes to identify candidate drugs that work to stop pseudoviruses in cells and that can move forward to the next phases of testing: live viruses and animal models.

“These funds will propel our investigations forward, but it’s important to realize how much work has gone on before—in AI, structural biology, drug development, cancer research, and many other fields—that makes this possible,” says Julien. “The reason we’re able to act quickly is because of the infrastructure that already exists and the long-term investments in basic science.”

Leveraging Biomedical Knowledge Graphs for COVID-19 Drug Repurposing Strategies Collaborators:
Jian Tang (Canada CIFAR AI Chair, Mila, HEC Montréal), William L. Hamilton (Canada CIFAR AI Chair, Mila, McGill University), Yoshua Bengio (Canada CIFAR AI Chair and co-director, CIFAR Learning in Machines & Brains program, Mila, Université de Montréal), Guy Wolf (Mila, Université de Montréal), Yue Li (Mila, McGill University)

AI-driven identification and validation of drug repurposing candidates to treat COVID-19 Collaborators:
Jean-Philippe Julien (CIFAR fellow, Molecular Architecture of Life program, University of Toronto), Costin Antonescu (Ryerson University), Cyclica Inc. (industry partner), Phoenox Pharma (industry partner)

line

AI research tool can help predict when, and how, to reopen the economy safely

The COVID-19 pandemic has closed the borders of dozens of countries around the world, and has forced people to practice social distancing and self-isolation. As the world adjusts to a new reality there are many questions and concerns about when, and how, to return to normal.

Canada CIFAR AI Chair Frank Wood and a team of researchers at the Programming Languages for Artificial Intelligence (PLAI) Research Group at the University of British Columbia (UBC) are investigating how to use existing software tools to automate parts of pandemic policy-making. Using existing epidemiological models, the team has shown how to use both a software tool and the techniques it developed to automate determination of which policies to put in place to reopen quickly and safely, such as hand-washing, social distancing, school closures, and their specific parameters. 

“Every model that we’ve seen implies that COVID-19 is going to be with us for some time.” says Wood, an associate professor at UBC and an associate member of Mila.  “Being locked up has put an immense strain on everyone.  Our aim is to help reopen the economy as quickly as possible while remaining safe, and the part we are well equipped to play in this is to inform epidemiologists and policy-makers about the most advanced tools and techniques available to do this now.”

“There was a groundswell of support and interest in doing good amongst members of my laboratory, rising almost to the level of mutiny,” says Wood.  The grant from CIFAR was a “welcome signal from the Canadian research funding landscape that this is a priority area.”

Wood is in ongoing discussions with Mila, MIT, Northeastern, Google, and various other teams. He is eager to expand collaboration with epidemiologists and policy-makers who are best equipped to use his software tools to support important policy decisions in the very near future.

Collaborators: 
Frank Wood (Canada CIFAR AI Chair, Mila, University of British Columbia), Benjamin Bloem-Reddy, Alexandre Bouchard, Trevor Campbell (University of British Columbia)