Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Master's Award: Canada Graduate Research Scholarships - Master's program

McGill University

$27,000.00

Benchmarking the Use of Graph Attention Neural Networks for Structure-Function Coupling Modelling

FY 2025-26

Summary
Structure-Function Coupling (SFC) characterizes the relationship between brain structure and brain function. Structure and function are both naturally expressed as graphs. As such, SFC can be modelled using Graph Neural Networks (GNNs). GNNs outperform other SFC modelling methods at the individual level. However, their practical value remains limited by major interpretability challenges, as the internal mechanisms by which GNNs integrate structure and function is difficult to interpret. Here, I propose using Graph Attention Neural Networks (GATs). Unlike GNNs, GATs assign attention coefficients to different graph nodes during the modelling process. These coefficients can be interpreted as revealing the important assigned by the GAT to different to different nodes during the modelling process. In this project, I will benchmark GATs to GANs suing the open-access and preprocess dataset Prevent AD. I will then modify the existing architecture of GATs to adopt a finer resolution of the structural of the structural and functional graphs, thus revealing which feature of each node is important to the modelling process. Finally, I will use the Virtual Brain in order to confirm my findings.
Program
Master's Award: Canada Graduate Research Scholarships - Master's program
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
CIHR Funding Decisions Database

As published

Id
572663
Name
Gregoire Philippe-Olivier
Region
Québec
Country
Canada
Orgtype
Institution Paid; Research Institution
Prcname2
Special Cases - Awards Programs
Projecttitle
Benchmarking the Use of Graph Attention Neural Networks for Structure-Function Coupling Modelling
Programtype2
Studentship Programs
Orgnameinp2
McGill University
Cihramount2
$27,000
Cihrequipment2
$0
Programname2
Master's Award: Canada Graduate Research Scholarships - Master's program
Keyworddelim
Diffusion-Weighed Magnetic Resonance Imaging; Functional Connectome; Functional Magnetic Resonance Imaging; Graph Attention Neural Networks; Regression; Structural Connectome; Structure-Function Coupling
Abstract
Structure-Function Coupling (SFC) characterizes the relationship between brain structure and brain function. Structure and function are both naturally expressed as graphs. As such, SFC can be modelled using Graph Neural Networks (GNNs). GNNs outperform other SFC modelling methods at the individual level. However, their practical value remains limited by major interpretability challenges, as the internal mechanisms by which GNNs integrate structure and function is difficult to interpret. Here, I propose using Graph Attention Neural Networks (GATs). Unlike GNNs, GATs assign attention coefficients to different graph nodes during the modelling process. These coefficients can be interpreted as revealing the important assigned by the GAT to different to different nodes during the modelling process. In this project, I will benchmark GATs to GANs suing the open-access and preprocess dataset Prevent AD. I will then modify the existing architecture of GATs to adopt a finer resolution of the structural of the structural and functional graphs, thus revealing which feature of each node is important to the modelling process. Finally, I will use the Virtual Brain in order to confirm my findings.
Competitiondate
202512
Approvedterm2
1 yr 0 mth
Pinamesdelim
Gregoire, Philippe-Olivier
Approvedterm
1 an 0 mois
Cihrcontribution2
~ $27,000~ $0~