T-CAIREM and the Data Sciences Institute Catalyst Grants support transformative data science research

by Chris Sasaki

The Data Sciences Institute (DSI) is pleased to announce the 2023 recipients of the annual DSI Catalyst Grant competition.

Catalyst Grants are awarded to multidisciplinary collaborative research teams focused on harnessing the transformative nature of the data sciences. The grants are given to teams working on the development of novel statistical or computational tools or the use of existing methodology in innovative ways to address questions of major societal importance and effect positive social change.

Two of this year’s Catalyst Grants are co-funded by the Temerty Centre for AI Research and Education in Medicine at the University (T-CAIREM). T-CAIREM seeks to establish new AI tools and data-driven projects that integrate clinical, translational and basic research into real-world applications.

“The recipients of 2023 grants are an illustration of the Institute’s mission to bring researchers together from across disciplines, divisions, campuses, as well as from the greater community, to address today’s critical challenges,” says Gary Bader, the Institute’s Associate Director, Research & Software. “It’s a truly remarkable list of impactful projects and an inspiring group of people.”

Thirteen interdisciplinary teams spanning all three campuses and external funding partners received grants (full list below), including three collaborations tackling diverse problems regarding harmful social media content, machine learning in healthcare and how marketing affects children’s health:

Reducing harmful content on social media through community-powered AI

Misinformation and hateful social media content has emerged as a critical threat, with several damaging impacts on our health, environment and society. While most social media platforms have taken measures to moderate and identify harmful content, and limit its spread, human moderators and AI algorithms often fail to identify it correctly and take proper actions. Historically marginalized groups are most affected by these failings as they have fewer representations among the human moderators, and their data are less available for the algorithms.

With their project, Syed Ishtiaque Ahmed from the Faculty of Arts & Science’s Department of Computer Science, Shion Guha from U of T’s Faculty of Information, and Shohini Bhattasali from University of Toronto Scarborough’s Department of Language Studies aim to improve the content moderation process by involving the communities affected by harmful or hateful content; and through a more pluralistic, contestability framework that allows multiple perspectives.

Ahmed, Guha, and Bhattasali will design, develop, deploy and evaluate the proposed system to address potentially Islamophobic and Sinophobic posts on Twitter in support of two Canadian non-profit organizations: the Chinese Canadian National Council for Social Justice (CCNC-SJ) and the Islam Unravelled Anti-Racism Initiative.

“Annotating data becomes challenging when the annotators are divided in their opinions” says Ahmed. “Democratically resolving this issue requires representing diverse values through the participation of different communities, which is currently absent in data science practices.

“This project addresses the issue by designing, developing and evaluating a pluralistic framework of justification and contestation in data science while working with two historically marginalized communities in Toronto.”

Tackling the labelling bottleneck in machine learning in healthcare

In many domains, labelling of objects — a fundamental step in machine learning — can be done by non-experts; for example, labelling can be as simple as drawing a box around a cyclist in an image.

“But in machine learning for healthcare (ML4HC) applications, the labeller needs to be a clinical domain expert,” says Sebastian Goodfellow, from the Department of Civil & Mineral Engineering, Faculty of Applied Science & Engineering.

“The time of clinician domain experts is scarce, and labelling becomes the rate-limiting step for ML4HC projects, preventing evaluation of the utility of machine learning in many medical applications. How to address this bottleneck is a knowledge gap in healthcare that leads to a translation gap, which our team will address.”

The goal of this project is to develop and evaluate two possible solutions to better understand this bottleneck. The first is a methodology for crowdsourcing labels from non-experts. The second is a novel framework for labelling medical waveform data by using a “human-in-the-loop,” semi-supervised learning pipeline and an interactive visualization approach.

The project is co-funded by T-CAIREM and the team comprises Goodfellow; and from the Hospital for Sick Children, Mjaye Mazwi, Translational Medicine Labs; Anica Bulic, Translational Medicine Labs; and Melissa McCradden, Genetics & Genome Biology Labs.

Says Goodfellow, “Our team is grateful to the DSI for this funding which will enable us to address an important challenge that affects many industries beyond healthcare. We’re excited to get started!”

Using deep learning and image recognition to measure child-directed food marketing

Childhood obesity and nutrition-related chronic disease are urgent global public health concerns. One of the factors contributing to the problem is that highly processed, energy-dense and nutrient-poor food is being marketed to children through tactics such as cartoon characters, toys and other fun enticements — all of which affect children’s attitudes, preferences and consumption behaviours.

But measuring child-directed marketing is time- and labour-intensive, requires in-depth training and validation, and is often subjective. As a result, there is a paucity of data with which to guide national and global legislation and policies aimed at protecting children from harmful industry practices.

The goal of this project is to develop new systems to capture food labels; develop methodologies using image recognition and deep learning technology to measure indicators of child-directed marketing on food and beverage packaging; and evaluate the relationship between child-directed marketing on food packaging, nutritional quality and price. These methodologies will enable evaluation of how child-directed food marketing may be perpetuating existing dietary and health inequities and whether policies are in fact reducing these disparities and protecting children’s right to health.

The team comprises experts from across four U of T academic divisions: Mary R. L’Abbé, Department of Nutritional Sciences, Temerty Faculty of Medicine; David Soberman, Joseph L. Rotman School of Management; Laura Rosella, Dalla Lana School of Public Health; and Steve Mann, Edward S. Rogers Sr. Department of Electrical & Computer Engineering, Faculty of Applied Science & Engineering.

“This project will help guide the development, implementation and evaluation of national and global food policies aimed at protecting children from harmful food industry marketing practices,” says L’Abbé. “Parliament is in the process of finalizing Bill C-252 to restrict the marketing of unhealthy foods to children and this grant can have a huge policy impact as part of our program on Food and Nutrition Policy for Population Health.” 

Launched in 2021, the DSI is the University of Toronto’s hub and incubator for data science research, training and partnerships, unifying research across the University, its affiliated institutes and external partners. 

Congratulations to all the 2023 DSI Catalyst Grant collaborative research teams!

A computational sociolinguistic approach for studying gender inequities in social media interactions

• Suzanne Stevenson (Department of Computer Science, Faculty of Arts & Science, U of T); Barend Beekhuizen (Department of Language Studies, University of Toronto Mississauga)

A high-throughput data and AI-driven mRNA transfection (HART) platform for immune cell engineering

• Bowen Li (Leslie Dan Faculty of Pharmacy, U of T); Bo Wang (Department of Laboratory Medicine and Pathobiology Temerty Faculty of Medicine, U of T)

Accelerating machine learning in healthcare: Solving the labelling bottleneck

• Project co-funded by T-CAIREM
• Sebastian Goodfellow (Department of Civil and Mineral Engineering, Faculty of Applied Science Engineering, U of T); Mjaye Mazwi (Translational Medicine Labs, The Hospital for Sick Children); Anica Bulic (Translational Medicine Labs, The Hospital for Sick Children); Melissa McCradden (Genetics & Genome Biology Labs, The Hospital for Sick Children)

Automating sedation state assessments

• Project co-funded by T-CAIREM
• Aaron Conway (Lawrence S. Bloomberg Faculty of Nursing, U of T); Babak Taati, Toronto Rehabilitation Institute, KITE, University Health Network); Sebastian Mafeld (Toronto General Hospital Research Institute, University Health Network)

DynaMELD and DynaCOMP: Using machine learning to revamp pre-and-post transplant care

• Rahul Krishnan (Department of Computer Science, Faculty of Arts & Science, U o f T); Mamatha Bhat (Toronto General Hospital Research Institute, University Health Network)

Inequality in childcare: The case of nannies in Canada

• Ito Peng (Department of Sociology, Faculty of Arts & Science, U of T); Monica Alexander (Department of Statistical Sciences, Faculty of Arts &Science, U of T)

Machine-learning-assisted screening of metallo-cyanines as light-absorbing and transport layers for organic and perovskite photovoltaics

• Oleksandr Voznyy (Department of Physical & Environmental Sciences, University of Toronto Scarborough); Timothy Bender (Department of Chemical Engineering and Applied Chemistry, Faculty of Applied Science & Engineering)

Providing data to improve representation by public officials

• Peter Loewen (Munk School of Global Affairs & Public Policy, Faculty of Arts & Science, U of T); Rohan Alexander (Faculty of Information, U of T); Aya Mitani (Dalla Lana School of Public Health, U of T); Elena Tuzhilina (Department of Statistical Sciences, Faculty of Arts & Science)

Something in the air: Is there an association between exposure to unconventional natural gas development (UNGD) and exacerbations of asthma in northeastern British Columbia?

• Élyse Caron-Beaudoin (Department of Health & Society, University of Toronto Scarborough); Marianne Hatzopoulou (Department of Civil & Mineral Engineering, Faculty of Applied Science & Engineering, U of T)

Spectroscopy by the millions: A fast, reproducible framework to yield chemical compositions of four million stars

• Joshua Speagle (Department of Statistical Sciences, Faculty of Arts & Science, U of T); Ting Li (David A. Dunlap Department of Astronomy & Astrophysics, Faculty of Arts & Science, U of T)

The rise of social media and the transformation of influence: Joining foundational sociological theory and data science to rethink influence in social systems

• Peter Marbach (Department of Computer Science, Faculty of Arts & Science, U of T; Vanina Leschziner (Department of Sociology, Faculty of Arts & Science, U of T; Daniel Silver (Department of Sociology, University of Toronto Scarborough)

Toward reducing harmful contents on social media with pluralistic justifications through community-powered AI

• Syed Ishtiaque Ahmed (Department of Computer Science, Faculty of Arts & Science, U of T); Shion Guha (Faculty of Information, U of T); Shohini Bhattasali (Department of Language Studies, University of Toronto Scarborough)

Using deep learning and image recognition to develop AI technology to measure child-directed marketing on food and beverage packaging and investigate the relationship between marketing, nutritional quality and price

• Mary R. L’Abbé (Department of Nutritional Sciences, Temerty Faculty of Medicine, U of T); David Soberman (Joseph L. Rotman School of Management, U of T); Laura Rosella (Dalla Lana School of Public Health, U of T); Steve Mann (Edward S. Rogers Sr. Department of Electrical & Computer Engineering, Faculty of Applied Science & Engineering, U of T)