TSpace

Preserve and Share Your Research

TSpace is a free and secure research repository established by University of Toronto Libraries to disseminate and preserve the scholarly record of University of Toronto.

Quick Access

View All Communities and Collections

Recent Submissions

Access status: Open Access ,
Designing an Ecological Restoration Solution with Novel Materials for a Circular Economy
(2026-05-28) Zhu, Xingyu; Krigstin, Sally
This study applies a circular economy framework to develop biodegradable substrates for ecological restoration using industrial byproducts. Recovered paper fiber was utilized as the primary matrix, with Biobinder® included as a binding component, and biochar and inactivated yeast incorporated as amendments. Eight formulations were evaluated under controlled laboratory conditions, with physical assessments focusing on water absorption under continuous immersion. Germination and root penetration were first observed using white clover (Trifolium repens). Based on these observations, three representative substrates were selected for further growth measurements (seedling height at days 15/30 and root length at day 30). Results showed that Biobinder® based substrates maintained strong structure despite lower water absorption and later became friable enough to support root penetration. While roots in pulp-only formulations tended to grow along the surface, biochar addition facilitated downward root development. Within the Biobinder® and biochar base substrate, the addition of inactivated yeast was found to significantly increase root length by day 30. Overall, the formulation combining Biobinder®, biochar, and yeast demonstrated the best performance. Despite limitations regarding short test duration and the lack of nutrient dynamic data, this study suggests that selected industrial byproducts have the potential to be transformed into functional restoration substrates. These findings provide experimental support for the development of low-cost, sustainable materials for large-scale ecological restoration.
Access status: Open Access ,
Minimal-Data Peptide Design and Accessible Protein Language Models for Protein Engineering
(2025-10) Bayat, Pouriya; Pardee, Keith; Pharmaceutical Sciences
Protein engineering is increasingly driven by machine learning, yet two practical barriers still limit broad adoption: (i) the scarcity of experimentally labelled data for new protein targets and (ii) the steep hardware demands of modern protein-language models (pLMs). This dissertation tackles both challenges through two core contributions. First, I introduce Minimal-Data → Maximal-Insight (MDMI), a structure-aware peptide-discovery pipeline that needs only a single round (~100 variants) of experimental screening. MDMI couples AlphaFold-Multimer complex prediction with hybrid statistical/physics scoring (SPServer + PyRosetta) to train a predictor, which then steers a genetic algorithm through sequence space to find novel sequences. As a case study, I applied MDMI to the split-GFP system, where the interaction between a 16-residue GFP11 peptide and its target GFP1-10 fragment reconstitutes fluorescence. MDMI successfully designed GFP11 peptides with over 50% sequence divergence from the wild type while preserving function, demonstrating the MDMI’s capacity to uncover non-obvious, high-diversity variants in data-limited scenarios. By decoupling peptide engineering from large training sets, MDMI offers an accessible strategy for laboratories with limited throughput.Second, I present Quantized Low-Rank Adaptation (QLoRA) for protein language models (pLMs), combining 4-bit weight quantization with low-rank adapters to enable efficient fine-tuning on lower cost and accessible GPUs. Across several pLMs (8 million–3 billion parameters) QLoRA cuts required training memory by an average 46.7%, and up to 90 % for the largest models, while retaining ≥ 90 % of baseline performance on regression tasks (i.e. fluorescence and stability datasets), secondary-structure classification, and de novo protein generation. Together, these methods establish a resource-conscious paradigm: MDMI extracts maximal design power from minimal data, and QLoRA delivers advanced pLMs to laboratories without specialized hardware. By uniting minimal-data modelling with hardware-efficient learning, this work broadens access to next-generation protein engineering and paves the way for rapid, distributed innovation in therapeutics, diagnostics, and biomanufacturing.
Access status: Open Access ,
Star formation across the scales
(2025-10) Khullar, Shivan; Matzner, Christopher; Murray, Norman; Astronomy and Astrophysics
The gas flows that form stars assemble at the scales of a few kiloparsecs and collapse into objects smaller than an astronomical unit. Different theories invoke different physical mechanisms to explain the inefficient nature of the star formation process -- some applicable on the global galactic scales, others on local cloud scales. In this thesis, we extend existing models at the cloud scale, test local models at the galactic scales and validate methods to increase resolution in simulations that bridge the gap in these two scales. First, we perform numerical simulations of self-gravitating turbulent flows in a patch of a molecular cloud. We study the physical origin of the gas density distribution (PDF) and explain the connection between the PDF and the dimensionless star formation rate. We also characterize the dependence of the various moving parts of the PDF on dimensionless cloud parameters. Second, we perform a numerical experiment at the galactic scale to understand the role of stellar feedback in the evolution of giant molecular cloud populations. We find that feedback drives turbulent motions, and test predictions of turbulence-regulated theories of star formation. Third, we test and validate a technique to increase resolution in Lagrangian simulations. We find that simulations where we increase resolution, compared to ones where we do not, exhibit systematic changes in the density field, masses and numbers of accreting stellar particles. However, the chaotic nature of star forming systems ultimately implies that simulations where we increase resolution do not sample a star formation history that might otherwise not exist. This thesis paves the way for future simulations that span the entire dynamical range of the star formation process.
Access status: Open Access ,
Multimodal Artificial Intelligence System to Enhance the Classification and Analysis of Diabetic Foot Ulcer Healing Conditions
(2025-10) Basiri, Reza; Popovic, Milos R; Khan, Shehroz S; Biomedical Engineering
Diabetic foot ulcers (DFUs) remain a leading cause of non-traumatic amputations and place a significant burden on healthcare systems. Effective treatment of these ulcers necessitates multidisciplinary approaches and complex standardized assessment methods. This thesis addresses this challenge by introducing an artificial intelligence-driven multimodal DFU classification framework that integrates structured metadata, optical imaging, thermal mapping, and depth sensing to improve DFU detection, localization, and healing phase prediction. This work introduces several significant contributions to DFU management.First, the creation of the Zivot dataset—the largest multimodal DFU collection to date—provides an unprecedented resource comprising approximately 10,000 annotated images with corresponding clinical metadata from 268 patients. Second, this research identifies and ranks 22 key clinical indicators significantly influencing healing phase classification through meticulous feature analysis, enabling accurate assessment without reliance on laboratory results. Third, the comprehensive evaluation of modality interactions reveals crucial insights into the complementary nature of different imaging techniques, establishing quantitative evidence for their relative contributions to wound assessment and planning. Finally, the Generative Adaptive Multimodal Attention Network development represents a novel architectural solution that optimally integrates these diverse data sources, achieving correct healing phase classification in more than 70% of cases. This is a substantial improvement over single-modality approaches and a clinically meaningful improvement that could enhance treatment planning and intervention timing in resource-constrained settings. Collectively, these advances establish a new paradigm for objective DFU assessment that enhances clinical decision-making and facilitates early intervention strategies to reduce amputation risk.
Access status: Open Access ,
Exploring the Neural Correlates of Body Perception Disturbances Using Visual Illusions in Complex Regional Pain Syndrome
(2025-10) Mockford, Matthew; Moayedi, Massieh; Dentistry
Complex regional pain syndrome (CRPS) is a debilitating chronic pain syndrome with poorlyunderstood mechanisms. Many individuals with CRPS experience distorted perceptions of their affected limb (neglect, body image mismatches, sensorimotor deficits), referred to as body perception disturbances (BPD). A visual hand morph illusion that modulates body image reduces BPDs and clinical pain in CRPS. Here, we investigated the neural correlates of this hand morph illusion using functional magnetic resonance imaging (fMRI) in individuals with CRPS. Forty- four participants (22 CRPS, 22 pain-free controls) completed the hand morph illusion while undergoing functional brain MRI, and provided ratings trial-by-trial on ownership, likeability, and pain of their hand. We found no main effect of the illusion compared to a sham control condition on pain intensity, limb ownership and likeability. We report differences in brain activation of body image-related regions in CRPS, compared to controls. These findings provide potential therapeutic targets for CRPS.