Western University Open Repository
Not a member yet
30251 research outputs found
Sort by
AI-Powered Drug Screening for Pin1 in Triple-Negative Breast Cancer Using a Pretrained Molecular Foundation Model
No full textTriple-negative breast cancer (TNBC) presents challenges due to its aggressiveness and lack of targeted therapies. Pin1, a prolyl isomerase involved in oncogenic signaling, has emerged as a therapeutic target. This study develops a scalable, AI-enabled virtual screening framework to identify Pin1 inhibitors using chemical language modeling and structure-based evaluation. MolFormer was fine-tuned on a bioactivity-labeled dataset to predict compound-induced changes in Pin1 expression. Ten million compounds were screened, and the top candidates were prioritized using a four-stage downstream pipeline consisting of activity prediction, molecular docking, molecular dynamics simulations, and drug-like property profiling. Seven compounds satisfied all criteria for predicted bioactivity, stable target engagement, and favorable pharmacokinetic properties. In silico validations, using Pin1 mutant and known Pin1 inhibitors, confirmed the framework’s reliability. The study illustrates the utility of pretrained chemical models in large-scale drug screening and establishes a reproducible strategy for integrating deep learning with structural and pharmacological evaluation
Territorializing Data for Indigenous Data Sovereignty
No full textThis thesis responds to a critical gap in Canadian research data management: the persistent lack of operational mechanisms to translate federal commitments to Indigenous data sovereignty (IDSov) into practice within research data repositories. While the TriAgency Research Data Management Policy, the United Nations Declaration on the Rights of Indigenous Peoples Act, and the Digital Research Alliance of Canada all affirm IDSov as a principle, meaningful operationalization remains limited. This dissertation asks: how can IDSov be operationalized within research data repositories?
The core contribution is a comprehensive framework for data territorialization—anchoring data within Indigenous jurisdictions and relational obligations through sociotechnical design. Drawing on Robert Sack's theory of human territoriality and Indigenous relational ontologies, the research conceptualizes territorialization as classification (encoding territorial boundaries in metadata), communication (surfacing Indigenous jurisdiction through discovery interfaces), and enforcement (operationalizing community protocols through access controls). These three functions operate cyclically to make data territories legible, actionable, and enforceable. Methodologically, the project introduces an integrated scalar-affordance framework that examines how territorial affordances operate across multiple interconnected scales—from individual user interactions to institutional policies, legal frameworks, and disciplinary discourses—revealing how colonial logics persist as reinforcing cycles across all levels
simultaneously. This methodology provides integrated diagnostic, design, and assessment functions for sovereignty-affirming repository redesign.
Through detailed analysis of metadata schemas, discovery architectures, and access control systems, the research specifies implementable territorial mechanisms: origin fields that preserve community-defined names and governance status, jurisdiction-based search enabling communities to find their own data, geospatial indexing anchoring data to place, and attribute-based access control enforcing community protocols. Three case applications demonstrate the framework's diagnostic and design utility: analysis of Canada's Federated Research Data Repository (FRDR/Lunaris) reveals how territorial affordances can exist technically but remain hidden or falsely perceived without proper signification; the Pima Indian Diabetes Dataset exemplifies systematic deterritorialization and its harms; and the Tsimshian Archaeological Spatial Archive demonstrates emerging territorial practice.
This thesis concludes with practical assessment guidance translating principles into repeatable sociotechnical patterns. It specifies implementation pathways for funders, platforms, and research organizations, and articulates how infrastructural decolonization—rebuilding research systems to support Indigenous sovereignty rather than perpetuate colonial extraction—is both necessary and achievable. By moving from recognition to realization, this work provides a blueprint for place-based data stewardship that respects Indigenous rights while supporting better research and stronger relationships with Indigenous nations for future generations.Danica Pawlick-Potts, 202
Thermochemical Valorization of Contaminated Plastics for Hydrogen Production
No full textPlastic waste from lead-acid batteries (LABs) is an increasingly significant yet under-addressed component in plastic waste and battery value chains. The casing and separators are heterogeneous mixtures of acrylonitrile-butadiene-styrene (ABS), polyethylene (PE), and fiberglass, heavily contaminated with halogens, sulfur, and metals, which make them intrinsically incompatible with conventional mechanical recycling and problematic for incineration and landfilling. This thesis investigates whether LAB-derived plastics can be converted into hydrogen-rich gas and carbonaceous solids by thermochemical routes while achieving robust immobilization of chlorine, bromine and sulfur, and whether such a process is technically and economically feasible when scaled to a 100 kg h-1 plant.
A real industrial LAB-plastic feedstock was characterized by ICP-MS and Ion Chromatography to quantify halogens, sulfur and metal content and distribution. Single and double-stage pyrolysis experiments were performed in a lab-scale continuous mechanically fluidized bed reactor coupled to a tubular flow reactor (TFR). In the single stage configuration, the influence of temperature (500-650°C) and in situ sorbents (CaO, Na2CO3) on product yields, gas composition and contaminants partitioning was quantified. In the double stage configuration, vapors generated at 500°C over a Na2CO3 bed were routed to a high-temperature tubular flow reactor operated between 890-1200°C. Experimentally, the double stage process increased hydrogen concentration in the product gas to ~87%-vol at 1200°C while maintaining gas yields near 50%-wt and generating 15%-wt secondary char. In situ Na2CO3 immobilized about 90% of the inlet chlorine, bromine and sulfur under the most favorable conditions. Fraction resolved TGA combined Fraser-Suzuki deconvolution with multi-step Sestrák-Berggren modelling to obtain kinetic triplets, which were implemented in Aspen HYSYS to construct reactor networks for a 100 kg h-1 plant and to support techno-economic analysis. The economic feasibility of the Aspen-based LAB-plastic plant achieves positive net present values at gas selling prices compatible with emerging low-carbon hydrogen markets when realistic gate fees are included.
Overall, the thesis demonstrates that LAB-derived plastics can be repositioned from a hazardous liability to a strategic feedstock for decentralized hydrogen and engineered carbon production, provided that double-stage pyrolysis with in situ sorbents and appropriate process integration is adopted.Maddalena Laghezza, 202
Trivariate Joint Modeling of Longitudinal, Recurrent, and Terminal Data in Clusters, with Applications to Lynch Syndrome Family Data
No full textTrivariate joint modeling for longitudinal measurements, recurrent events and a terminal event in clustered data has attracted increasing interest in medical studies. For example, families with Lynch Syndrome (LS) face elevated colorectal cancer (CRC) risk, where the number of polyps detected during colonoscopy and the frequency of screenings could influence CRC risk at both individual and family levels. Detecting and removing polyps can potentially delay or prevent CRC onset. To assess how screening affect polyp detection and subsequent CRC risk, this thesis proposes a clustered trivariate joint model for longitudinal counts, recurrent events, and a terminal event. The proposed model facilitates zero-inflation and over-dispersion in longitudinal counts, and invokes subject-specific and family-specific random effects to capture dependencies. The primary goal is to develop a framework for analyzing clustered longitudinal count data with recurrent events and a terminal event, enabling comprehensive analysis and dynamic predictions, while accounting for measurement error in longitudinal count data.
The first project introduces a trivariate joint model treating CRC occurrence as a terminal event, colonoscopy visits as recurrent events, and polyps counts as zero-inflated, over-dispersed longitudinal count data. Comparisons with existing trivariate and bivariate joint models demonstrate superior fit and highlight the importance of familial clustering, revealing heterogeneity in polyp detection and CRC risk that informs targeted screening strategies.
The second project focuses on dynamic prediction of CRC risk and polyp counts by incorporating individual and familial histories. By accounting for familial correlations and family history, the proposed model provides more accurate predictions than other trivariate joint models, particularly when extensive histories are accommodated.
The third project addresses measurement error in longitudinal count data, a common but often ignored issue. To mitigate the measurement error effect, a regression calibration method is introduced to improve inference. Simulation studies and real data analysis demonstrate that this approach enhances parameter estimation and predictive accuracy when longitudinal counts are subject to measurement error.
This research develops a general trivariate joint modeling framework for clustered data involving longitudinal counts, recurrent events, and a terminal event, offering improved dynamic prediction and methods for studying disease progression.Jingwei Lu, 202
Functional anatomy of the quadriceps femoris: Influence of muscle, sex, and fatigue on motor unit behaviour
No full textOur understanding of human physiology, particularly the neuromuscular system, remains incomplete due to the scarcity of data concerning anatomically deep muscles and the historical underrepresentation of female participants. This thesis addressed these critical gaps by investigating muscle- and sex-related differences in motor unit (MU) behaviour within the quadriceps femoris. Across the three experimental chapters, direct intramuscular electromyography recordings of MU firing rates were made from the anatomically deep vastus intermedius (VI) and the more superficial vastus lateralis (VL) during isometric knee extension contractions.
Study 1 aimed to characterize the neuromuscular control of the VI in relation to the well-studied VL across a full force range of voluntary contractions. Contrary to expectations based on architectural and biomechanical differences, the results demonstrated that MU firing rates were similar between the VI and VL. This indicates that despite their distinct anatomical structures, the fundamental neural activation strategies of these two major torque-generating synergists are comparable at the MU level, implying that any variations in their relative contribution to knee extension torque may be governed by biomechanics rather than divergent MU rate coding behaviour.
Study 2 focused on sex-related control of the neuromuscular system by examining MU firing rates and contractile properties across a range of voluntary forces including at maximum intensity. Females had higher firing rates at submaximal intensities compared to males, but there were no differences at maximum voluntary contraction (MVC) levels. Furthermore, measures of electrically-induced isometric contractile properties, particularly when normalized to maximum, were similar between sexes. These findings indicate that force modulation through rate coding differs between males and females during submaximal contractions in two large muscles of the quadriceps. This difference is likely driven by intrinsic motor neuron excitability rather than intrinsic muscle properties such as contractile speed.
Study 3 investigated the combined influence of muscle and sex during a sustained maximal fatiguing contraction (one-minute MVC) and subsequent recovery period. The study aimed to parse out any differences when the neuromuscular system was maximally stressed. Results showed that both muscle (VI vs. VL) and sex (males vs. females) exhibited similar rate coding behaviour from the fatiguing tasks and during five minutes of recovery. These findings indicate that comparable neural control strategies, as expressed by firing rate changes, are used in these muscles and between sexes when the neuromuscular system is stressed causing substantial muscle fatigue.
The three studies provide foundational data on the rate coding of deeper anatomical muscles and their behaviour within a synergistic group, as well as advancing our understanding of the sex-related control of the neuromuscular system both submaximally and under maximal stress
Oxygen and Oxidative Stress in the Hibernating 13-Lined Ground Squirrel
No full textN/AThe hibernating 13-lined ground squirrel (TLGS), Ictidomys tridecemlineatus, experiences dozens of cycles between extreme metabolic suppression in torpor and arousals to interbout euthermia (IBE). As metabolism, heart rate, and body temperature rise, likely O2 fluctuations should disrupt mitochondrial reduction and oxidation (redox) balance, increasing oxidative stress. In non-hibernators, ischemia (no blood flow) drives succinate accumulation and reduces the ubiquinone pool; upon reperfusion, this imbalance promotes reverse electron transport (RET) and reactive oxygen species (ROS) production at complex I, causing significant oxidative damage. Hibernators demonstrate enhanced ischemia–reperfusion tolerance, with less oxidative damage and cellular dysfunction than non-hibernators. In isolated liver mitochondria from TLGS, succinate oxidation is suppressed during torpor and is slowly reactivated upon arousal. Metabolic suppression may also minimize ROS production, lowering antioxidant demand and promoting nonenzymatic antioxidant accumulation. I hypothesized that metabolic suppression protects hibernators from oxidative stress and predicted that, throughout torpor and early arousal, suppressed succinate oxidation and antioxidant accumulation would limit rates of ROS production and oxidative damage in the TLGS. With pulse oximetry, I found that carotid artery O2 saturation (SaO2) increased 2.6-fold during 2–3 hours of arousal, indicative of rapid reoxygenation. Liver metabolomics showed that, despite suppressed succinate oxidation, succinate accumulated 39-fold throughout torpor and early arousal, indicative of reversed succinate dehydrogenase activity. Concurrently, glutathione precursors and amino acids with independent antioxidant capacity accumulated. This suggests hypoxia in torpor and an increased potential for ROS production upon arousal; however, the accrued antioxidant defence may quench generated ROS. In isolated liver mitochondria, succinate was 15 % higher in torpor compared to summer, and antioxidants did not change among states. In anoxia, only summer mitochondria accumulated succinate, indicating enhanced redox control during hibernation. With complex I and RET inhibition by rotenone, torpid ROS production declined, suggesting that in vivo suppression of complex II similarly protects TLGS liver mitochondria from RET-associated ROS production. Torpid mitochondria maintained respiratory function following anoxia–reoxygenation, but in IBE and summer, O₂ consumption declined. Hibernators are a model of reoxygenation tolerance, where metabolic suppression confers oxidative protection.Brynne Duffy, 202
Understanding Learning in Non-Stationary Environments
No full textDomain Generalization (DG) aims to train models that can perform well on unseen target domains by learning from multiple related sources. Unlike classical empirical risk minimization, which assumes training and test data come from an Independent and Identical Distribution (i.i.d.), DG explicitly tackles distribution shifts by seeking shared knowledge across domains. However, they often rely on simplified assumptions such as stationary and independent domain sampling. In practice, data evolve over time or space, making domains rarely static. As a result, DG methods trained under non-stationary conditions frequently fail, and empirical results show that ignoring temporal evolution can degrade predictive performance.
To bridge this gap, we introduce Evolving Domain Generalization (EDG), which characterizes domains that evolve temporally driven by latent dynamic factors. While standard DG addresses static distribution shifts, EDG is aimed at capturing the evolving pattern, the latent structure driving domain evolutions (see Figure 1.2). Uncovering and leveraging this pattern allows us to generalize effectively in non-stationary environments.
This thesis develops a principled EDG framework by answering four essential questions that arise when domain evolution is present:
1. How should domain evolution be defined? (Chapter 2): We define domain evolution as the action of a latent dynamical operator governing distributional change across domains. Based on this definition, we propose Temporal Koopman Networks (TKNets) to embed evolving distributions into a Koopman-governed latent space, linearizing non-linear dynamics and enabling principled alignment across sequential domains.
2. How can a future evolving domain be anticipated during training? (Chapter 3): Building on the need to anticipate future domains, we introduce Domain Directional Augmentation (DDA). This meta-learning approach acts as a generative engine for the framework, utilizing an attention-based mechanism to simulate feature augmentations along the evolving domain directions. It enables the model to proactively adapt to unseen distribution changes, moving beyond the constraints of static data augmentation.
3. How can evolving patterns be captured from sparse observations? (Chapter 4): To address the real-world challenge of data sparsity, we propose the Infinitely Fine-Grained Evolving Trajectory (IFGET) model. By modeling the evolving pattern as a Stochastic Differential Equation (SDE), this method bridges discrete, sparse snapshots to reconstruct a continuous evolutionary trajectory, ensuring the capture of evolving patterns even when temporal observations are sparse.
4. When do evolving patterns exist? (Chapter 5): Finally, to complete the framework, we provide the diagnostic tools necessary to detect and measure these patterns. We introduce EvoRate (and its optimal-transport variant EvoRateW), a metric to quantify the existence and intensity of domain evolution. This allows us to explicitly verify the existence of evolving patterns and determine the applicability of EDG methods.
Together, these components establish a framework that characterizes the latent structure of domain evolution and provides the necessary tools to simulate, recover, and measure it in complex dynamic environments
An expert-driven consensus framework for the study of potentially morally injurious events and their impacts: findings from an e-Delphi study
No full textBackground: Moral injury (MI) refers to the profound psychosocial, spiritual and behavioural
impacts of exposure to potentially morally injurious events (PMIEs). Despite growing recognition
of MI across military and civilian contexts, definitional clarity surrounding PMIEs remains limited.
Objective: This study applied e-Delphi methodology to generate consensus on the defining
features of PMIEs and their impacts among an interdisciplinary panel of MI experts.
Method: The panel first provided narrative responses to open-ended questions on defining PMIEs.
These were refined into 63 Likert statements. Experts rated agreement on these and completed
card-sort and ranking exercises.
Results: Statements addressed eight themes: exposure, transgressive acts, consequences of PMIEs,
trauma vs. PMIEs, moral agency, betrayal, subjectivity, and high-stakes. Consensus (≥80%
agreement) was reached for 55% of all statements. Consensus for card-sort and ranking exercises
was also observed, pertaining to etiological mechanisms of MI, and risk and protective factors.
Themes with the highest levels of consensus included exposure and transgressive acts, while
moderate consensus was achieved on PMIE consequences and comparisons to trauma. Lower
consensus emerged around moral agency, betrayal, high-stakes and subjectivity, substantiating
these as areas of ongoing debate.
Conclusions: This study clarifies key definitional features of PMIEs and their impacts, with findings
organised into a consensus framework for the future study of PMIEs. Findings highlight the need for
empirical testing of proposed features and areas of debate, integration with emerging trauma
frameworks, and culturally inclusive approache
Digital Resilience in Social Media Feminist Activism: Reactance Theory Applied to Weibo and Zhihu
No full textThis article is part of the issue “Digital Resilience Within a Hypermediated Polycrisis” edited by
Marc Esteve Del Valle (University of Groningen), Ansgard Heinrich (University of Groningen), and Anabel
Quan‐Haase (Western University).Past studies have shown the value of social media for feminist activism in China. Yet, activists encounter strict censorship, negatively impacting their mobilization efforts. Existing studies have documented the strategies activists use to circumvent censorship by analyzing digital trace data but have not yet examined their censorship experiences. To fill this gap, the present study draws on reactance theory to investigate the experiences of social media feminist activists in China through 19 in-depth interviews. Following calls to examine digital resilience in the era of polycrisis, this study also contributes to rethinking digital resilience as not only resistance to censorship, but as an adaptive capacity to maintain agency and continuity in activism. We conducted a cross-platform comparison contrasting activists’ censorship experiences across Weibo and Zhihu. We found a hierarchy of concerns underlying censorship mechanisms. We identified five types of cognitive reactance: ambiguity, disagreement, unfairness, believing in a lack of control, and critical questioning of the positive energy motto. Affective reactance manifested through feelings of anger and irritation toward haphazard censorship enforcement. Digital resilience was visible in both cognitive and affective reactance, which motivated participants to restore their freedom. Participants used two types of direct means to regain their lost freedom: seeking and disseminating censored information. A few participants engaged in indirect restoration by reinterpreting the state’s motto of positive energy. The findings suggest activists developed different forms of digital resilience on Zhihu and Weibo that reflect unique platform affordances and regulations. We outline implications for reactance theory and future research
Caregiver-Implemented Interventions in Pediatric Speech-Language Pathology: Lessons Learned from Practice-Based Research
No full textCaregiver-implemented interventions are often used by speech-language pathologists to support children’s early language development by empowering caregivers to facilitate children’s language learning in everyday contexts. The Hanen Centre is a globally recognized organization that trains speech-language pathologists to deliver caregiver- implemented interventions. Two Hanen programs focus on supporting the language development of children with more general communication needs – Target Word and Learning Language and Loving It. The Target Word program aims to improve the language and communicative participation skills of children who are late-to-talk. The Learning Language and Loving It program engages early childhood educators in supporting children’s language learning in childcare contexts. This dissertation explored the effectiveness, feasibility, and cultural appropriateness of these programs in applied contexts.
Study 1 evaluated the effectiveness of the virtual Target Word program. A constrained longitudinal data analysis approach was used to examine differences in vocabulary, morphology, phonology and communicative participation for children in immediate and waitlist control groups. Clinically meaningful gains in communicative participation and statistically significant gains in phonology were associated with families’ program involvement, however most other scores in expressive vocabulary, morphosyntax, and communicative participation were not statistically different between groups. Study 2 assessed parent-child interaction quality during video-recorded play-based interactions for the same families who participated in Study 1. Parent-child interactions were coded using the Parenting Interactions with Children: Checklist of Observations Linked to Outcomes (PICCOLO) and data were analyzed using constrained longitudinal data analysis. No group differences were observed at any point. Study 3 aimed to understand Red River Métis Early Childhood Educators’ perceptions of the feasibility and the cultural appropriateness of the strategies recommended in the Learning Language and Loving It program. A think aloud interview approach was used to capture participants’ views. A thematic analysis revealed three themes related to the feasibility of program strategies, and two themes identified suggested cultural adaptations and considerations for Red River Métis childcare centres.
This dissertation provides insight into the effectiveness of caregiver-implemented interventions and key cultural considerations for those working with Red River Métis children and their caregivers. Practical implications for research and practice in pediatric speech-language pathology are also identified.Kathryn Hatherly, 202