Leonid Kholkine

• 17/01/2025
  
• 15.00 uur
• Locatie: Stadscampus, Hof Van Liere, F. De Tassiszaal
• Promotoren: Tim Verdonck & Steven Latré
• Departement Informatica

Abstract

While much of Machine Learning research relies on controlled "toy datasets," real-world data and applications are significantly more unpredictable, introducing unique challenges in variability, scalability, and practical implementation. This thesis examines the practical applications of Machine Learning in real-life scenarios, focusing on personalized recipe recommendations and sports analytics. By addressing these complexities, it bridges the gap between theoretical advancements and practical deployment, providing insights and solutions for real-world use cases.

In the retail domain, a graph-based recommendation system models relationships between users, recipes, and ingredients to deliver personalized, actionable suggestions. The system addresses issues such as popularity bias and the cold-start problem, achieving a 40% improvement in accuracy over traditional methods. The framework also enhances diversity, reduces bias, and provides explainability by identifying key features driving recommendations, offering a practical solution for improving user engagement and sales.

In sports, Machine Learning techniques predict cycling race outcomes using a learn-to-rank approach. This novel application explains predictions using SHAP values, highlighting race-specific factors that influence performance. These insights provide a tool for pre-race analysis and data-driven strategy development. Future enhancements, including probabilistic models and simulations, could extend predictions across seasons and multi-day events.

A data-driven analysis of professional cycling identifies the age of peak performance across athlete specializations, including sprinters, all-rounders, and general classification riders. By combining classification and clustering techniques, the research offers actionable insights for athlete career planning. Personalized models could refine these findings further by tailoring predictions to individual trajectories.

The development of a low-cost, real-time player tracking system for field hockey demonstrates the potential of affordable technology to democratize sports analytics. Using a simple GoPro setup, the system achieves 93.8% accuracy in tracking players, providing underrepresented sports with access to data-driven insights. Further refinements could improve tracking performance, especially in high-traffic scenarios.

By addressing real-world challenges and proposing future advancements, this research highlights the transformative potential of Machine Learning in retail and sports. From improving recommendations to optimizing athletic performance and engagement, it offers a practical roadmap for deploying advanced algorithms in diverse, impactful applications.

Jonas Torfs

• 15/01/2025
  
• 17.00 uur
• Locatie: Campus Drie Eiken, O.05
• Promotoren: Marcel Eens & Nicky Staes
• Departement Biologie

Abstract

The animal gut houses a diverse community of bacteria, called the gut bacteriome, that performs essential functions in physiology, metabolism, and overall health. As such, it allows its host to respond to local environmental conditions, directly influencing host fitness. Despite the well-established role of the gut bacteriome in host biology, the forces shaping its composition within and between individuals of the same species remain poorly understood. 

In this thesis, I explored the gut bacteriome and its associated exposome and host-intrinsic factors in one of humans’ closest living relatives, the bonobo. In the first section of my thesis, I focus on addressing certain knowledge gaps in potential covariates of the bonobo gut bacteriome. First, I validated a method to assess body condition in bonobos and showed that it is heritable, and independent from dietary patterns. In another study, I examined the predictors of an individual’s position in the grooming network using a decade of data from zoo-housed bonobos. Age and rearing history impacted grooming behaviour, often in sex-specific manners, which fosters a better understanding of the link between grooming and the gut bacteriome. Finally, I showed that respiratory disease spreads following the proximity network in a zoo-housed group of bonobos.

The second section focused on mapping the bonobo gut bacteriome and its associated factors. I collected faecal samples with associated data from zoo-housed and wild bonobos and subjected them to 16S rRNA amplicon sequencing to determine gut bacterial profiles. Zoo-housed and wild bonobos differed remarkably in their gut bacteriome diversity and composition. Within zoo-housed individuals, lifestyle factors such as diet, early-life adversity, and medication use influenced bacteriome structure. On a dyadic level, social contact and environmental sharing were strongly correlated with gut bacteriome similarity, overshadowing maternal and genetic effects. 

Finally, I focused my attention on ageing patterns in the bonobo gut bacteriome, showing that across lifestyles (zoo vs wild) diversity increased with age while the core bacteriome shrank. Interestingly, ageing zoo-housed bonobos became more unique in their gut bacterial composition while this pattern was not observed in the wild, mirroring the human pattern. This points towards conserved aging processes in the hominoid gut bacteriome. In summary, this thesis offers key insights into the mechanisms through which the exposome influences gut bacterial communities in great ape species of which the gut bacteriome was not yet adequately characterized.

Adwine Vanslembrouck

• 13/01/2025
  
• 16.00 uur
• Locatie: ITM, Campus Rochus, Aula Janssens, Sint-Rochusstraat 43, 2000 Antwerpen
• ​Online Doctoraatsverdediging
• Promotoren: Herwig Leirs & Ruth Müller (ITM)
• Departement Biologie

Abstract

Invasive Aedes mosquitoes, such as the tiger mosquito (A. albopictus) and the Asian bush mosquito (A. japonicus), are known vectors for various viruses. One of the most significant consequences of these invasive mosquito infestations in Europe is the outbreak of mosquito-borne diseases such as dengue and chikungunya. These invasive Aedes mosquitoes are known to spread to new areas due to climate change and globalization. Additionally, anthropization to the environment, such as urbanization and deforestation, are important drivers of their dispersal. These human interventions lead to biotic homogenization, resulting in reduced biodiversity. This results in a reduced competition from native mosquitoes and a decrease in predators that could limit the spread and establishment of invasive mosquitoes. Furthermore, there is an increase in insecticide resistance, making it more difficult to control mosquito populations. As a result, there is a growing need for alternative control methods, such as biological control.

This thesis investigates the ecological interactions, competition dynamics, and implications for vector control strategies, with a particular focus on the invasive Aedes mosquitoes and native Culex species. The research covers interspecific larval competition, thermophilic preferences, and arboviral infection risks, demonstrating how larval competition among mosquito species affects their susceptibility for arboviruses. We found that Aedes albopictus larvae benefit from competing with Culex hortensis, potentially amplifying arbovirus transmission risks. Additionally, larval interactions among Aedes albopictus, Aedes japonicus, and Culex pipiens reveal metabolic and behavioral adaptations that increase susceptibility to arboviruses like chikungunya and West Nile virus.

The insecticide resistance status in European Culex pipiens populations highlights the limitations of conventional mosquito control methods. Resistance to multiple insecticides, also found in Belgium, points out the urgent need for innovative alternative approaches, including biological control. In this research, the role of native predatory diving beetles as natural regulators of mosquito populations in Belgium was investigated, identifying Agabus bipustulatus as a highly effective predator for Aedes albopictus.

In sum, this thesis emphasizes the importance of ecological interactions in understanding mosquito-borne disease risk and in developing sustainable mosquito control strategies. By integrating insights from larval competition, arboviral risks, and alternative control methods, it contributes to a broader understanding of vector ecology and offers novel approaches to mitigate the public health impacts of invasive mosquitoes in Europe. ​ ​

Ignace Pelckmans

• 09/01/2025
  
• 17.00 uur
• Locatie: Campus Middelheim, A.143
• Promotor: Stijn Temmerman
• Departement Biologie

Abstract

River deltas and estuaries are densely populated regions but face coastal flood risks due to extreme sea level events, particularly as climate change leads to rising sea levels and more instense storms. Nature-based strategies, such as the conservation and restoration of mangroves, are increasingly considered a cost-effective and sustainable addition to traditional flood defences. Cities such as Guayaquil (Ecuador), Ho Chi Minh (Vietnam) and Khulna (Bangladesh) all house millions of people but despite their growing exposure to coastal floods, the mangroves in between these cities and the sea have been largely lost to aquaculture. Yet, how the largescale conversion from mangrove to human land use affects the propagation of flood waves is still unknown. This thesis attempts to fill that knowledge gap by elaborating on how extreme sea levels are heightened or lowered when propagating through a tropical delta, and how this depends on the spatial properties of the complex network of mangroves, estuarine channels and human land use that is intrinsic to most tropical deltas. Hydrodynamic models allow the study of flood wave propagation over the tens to hundreds of kilometres that tropical river deltas typically cover. In this thesis, we present the setup, calibration and evaluation of a hydrodynamic model of the Guayas delta, Ecuador. The model demonstrated that the current mangroves in the delta effectively attenuate incoming extreme sea levels related to El Niño events and prevent the increase in sea level to amplify upstream. Furthermore, we show that the spatial locations of mangroves and aquaculture have a large impact on the attenuation of extreme sea levels with the strongest reduction in case mangroves are uniformly spread over the delta or concentrated upstream. During a field campaign, we measured high water level attenuation in a tropical Rhizophora forest, where aerial roots are several metres high, revealing attenuation rates up to 46 ± 9.8 cm/km, which are the highest attenuation rates ever recorded in a mangrove forest. With this thesis, we show how mangrove ecosystem properties and their spatial location within deltas affect the reduction of extreme water levels when propagating through the complex web of mangroves, channels and aquaculture, which is intrinsic to many tropical river deltas. Human infrastructure in the deltaic plain is often essential for the economic development of these regions, yet this thesis elaborates on where conversion from mangroves to human land use has minor effects on the reduction of high water levels.