Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

Simon Fraser University

$59,000.00

Transfer learning and causal models for biomedical data

FY 2024-25

Summary
The objective of this proposal is to investigate machine learning methods for biomedical datasets that play a crucial role in many scientific domains. For example, in computational chemistry, machine learning models to predict the interaction between two molecules are trained on small datasets of interactions observed in expensive wet lab experiments. As another example, precision medicine relies on machine learning methods for personalized diagnosis and prognosis of patients. Available training datasets contain just hundreds of patients but each with mutation, expression and possibly other datatypes for tens of thousands of genes. These scientific domains come with the requirement of explanations to justify a prediction, i.e. diagnosis or treatment recommendation, and the need for causal knowledge that may represent mechanisms. We will adopt a framework based on (deep) neural networks. The architecture of the neural network will be chosen to fit the type of input data, e.g. graph neural networks for molecules in graph representation and feedforward neural networks for tabular data. As one approach to deal with the small size of training datasets, we will investigate various paradigms of transfer learning that transfer knowledge between different datasets (domains). To address the need for causal knowledge, we will explore methods for discovering causal knowledge from observational data. Causal models are also expected to transfer better between domains and to be more useful for producing explanations (because causal features are likely to correspond to mechanisms) than correlational models which are the standard in machine learning. Our first main research direction is transfer learning. We will explore two major paradigms of transfer learning: domain adaptation, which assumes a labeled source dataset and an unlabeled target dataset, and domain generalization, which assumes that at training time multiple source datasets but no target dataset are available. Our second main research direction is causal discovery from observational data. Causal knowledge will be represented using structural causal models (SCMs), i.e. directed acyclic graphs among the variables of interest together with a set of structural equations that govern the dependencies between variables. Our longstanding collaborations with academic and industry researchers, in particular those at the Vancouver Prostate Centre (a leading cancer research centre) and at Terramera Inc. (a successful startup company in the area of precision agriculture), will ensure that the developed machine learning methods will address the requirements of real-life scenarios and will have the potential to be translated into practical tools.
Program
Discovery Grants Program - Individual
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-05136
Name nom
Ester, Martin
Department département
Computing Science
Organization id
5
Institution établissement
Simon Fraser University
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
59000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2700
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Information technology
Research subject fr
Technologie de l'information
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Transfer learning and causal models for biomedical data
Keywords
machine learning; transfer learning; causal discovery and inference; biomedical applications; precision medicine
Application summary
The objective of this proposal is to investigate machine learning methods for biomedical datasets that play a crucial role in many scientific domains. For example, in computational chemistry, machine learning models to predict the interaction between two molecules are trained on small datasets of interactions observed in expensive wet lab experiments. As another example, precision medicine relies on machine learning methods for personalized diagnosis and prognosis of patients. Available training datasets contain just hundreds of patients but each with mutation, expression and possibly other datatypes for tens of thousands of genes. These scientific domains come with the requirement of explanations to justify a prediction, i.e. diagnosis or treatment recommendation, and the need for causal knowledge that may represent mechanisms. We will adopt a framework based on (deep) neural networks. The architecture of the neural network will be chosen to fit the type of input data, e.g. graph neural networks for molecules in graph representation and feedforward neural networks for tabular data. As one approach to deal with the small size of training datasets, we will investigate various paradigms of transfer learning that transfer knowledge between different datasets (domains). To address the need for causal knowledge, we will explore methods for discovering causal knowledge from observational data. Causal models are also expected to transfer better between domains and to be more useful for producing explanations (because causal features are likely to correspond to mechanisms) than correlational models which are the standard in machine learning. Our first main research direction is transfer learning. We will explore two major paradigms of transfer learning: domain adaptation, which assumes a labeled source dataset and an unlabeled target dataset, and domain generalization, which assumes that at training time multiple source datasets but no target dataset are available. Our second main research direction is causal discovery from observational data. Causal knowledge will be represented using structural causal models (SCMs), i.e. directed acyclic graphs among the variables of interest together with a set of structural equations that govern the dependencies between variables. Our longstanding collaborations with academic and industry researchers, in particular those at the Vancouver Prostate Centre (a leading cancer research centre) and at Terramera Inc. (a successful startup company in the area of precision agriculture), will ensure that the developed machine learning methods will address the requirements of real-life scenarios and will have the potential to be translated into practical tools.