Graph Attention Mechanisms for Modeling Pathway-Level Importance from Gene Expression

High-throughput sequencing technologies have revolutionized the analysis and profiling of gene expression data, enabling a comprehensive understanding of the complex mechanisms underlying molecular biology. However, the high dimensionality and noise inherent in gene expression data pose significant challenges for biomarker discovery and molecular subtype classification in cancer. While deep learning models have shown promise in capturing complex transcriptomic patterns, their black-box nature limits their interpretability for downstream biological insight and analysis. Given that cancer phenotypes may arise from the dysregulation of gene networks, incorporating pathway-level biological knowledge directly into model design offers a principled approach to improve robustness, interpretability, and predictive performance. This work presents a pathway-aware variational autoencoder which integrates curated gene-pathway structure into latent representation learning. This includes the integration of gene interactions through graph neural networks, and attentionbased hierarchical pooling from genes to pathways to latent sample representations. Across experimental trials, this architecture demonstrated stable training which had matched, if not improved, reconstruction and classification performance relative to gene-only baselines. Moreover, the use of attention mechanisms enabled the extraction of task-dependent gene importance scores, providing intrinsic interpretability to the model’s predictions. Conclusively, this work demonstrates that pathway-aware representation learning unifies predictive performance with biologically grounded interpretability, enabling robust systems-level biomarker discovery and molecular subtype classification in cancer.

Name thesis:degree_name

M.Sc. Biological Sciences

Level thesis:degree_level

Master

Discipline thesis:degree_discipline

Faculty of Mathematics and Science

Department dc:contributor.department

Department of Biological Sciences

Grantor dc:publisher

Brock University

Year dc:date.issued

2026

Author dc:creator

• Villena, Marcus

Advisor dc:contributor.advisor

• Yifeng, Li

• NATURAL SCIENCES::Biology::Cell and molecular biology
• TECHNOLOGY::Information technology::Computer science::Computer science

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/10464/20376

OAI identifier oai:identifier

oai:brocku.scholaris.ca:10464/20376