Faculty of Science (ETDs)

Permanent URI for this communityhttps://hdl.handle.net/10539/37933

For queries relating to content and technical issues, please contact IR specialists via this email address : openscholarship.library@wits.ac.za, Tel: 011 717 4652 or 011 717 1954

Browse

Search Results

Now showing 1 - 10 of 77

Towards Lifelong Reinforcement Learning through Temporal Logics and Zero-Shot Composition
(2024-10) Tasse, Geraud Nangue; Rosman, Benjamin; James, Steven
This thesis addresses the fundamental challenge of creating agents capable of solving a wide range of tasks in their environments, akin to human capabilities. For such agents to be truly useful and be capable of assisting humans in our day-to-day lives, we identify three key abilities that general purpose agents should have: Flexibility, Instructability, and Reliability (FIRe). Flexibility refers to the ability of agents to adapt to various tasks with minimal learning; instructability involves the capacity for agents to understand and execute task specifications provided by humans in a comprehensible manner; and reliability entails agents’ ability to solve tasks safely and effectively with theoretical guarantees on their behavior. To build such agents, reinforcement learning (RL) is the framework of choice given that it is the only one that models the agent-environment interaction. It is also particularly promising since it has shown remarkable success in recent years in various domains—including gaming, scientific research, and robotic control. However, prevailing RL methods often fall short of the FIRe desiderata. They typically exhibit poor sample efficiency, demanding millions of environment interactions to learn optimal behaviors. Task specification relies heavily on hand-designed reward functions, posing challenges for non-experts in defining tasks. Moreover, these methods tend to specialize in single tasks, lacking guarantees on the broader adaptability and behavior robustness desired for lifelong agents that need solve multiple tasks. Clearly, the regular RL framework is not enough, and does not capture important aspects of what makes humans so general—such as the use of language to specify and understand tasks. To address these shortcomings, we propose a principled framework for the logical composition of arbitrary tasks in an environment, and introduce a novel knowledge representation called World Value Functions (WVFs) that will enable agents to solve arbitrary tasks specified using language. The use of logical composition is inspired by the fact that all formal languages are built upon the rules of propositional logics. Hence, if we want agents that understand tasks specified in any formal language, we must define what it means to apply the usual logic operators (conjunction, disjunction, and negation) over tasks. The introduction of WVFs is inspired by the fact that humans seem to always seek general knowledge about how to achieve a variety of goals in their environment, irrespective of the specific task they are learning. Our main contributions include: (i) Instructable agents: We formalize the logical composition of arbitrary tasks in potentially stochastic environments, and ensure that task compositions lead to rewards minimising undesired behaviors. (ii) Flexible agents: We introduce WVFs as a new objective for RL agents, enabling them to solve a variety of tasks in their environment. Additionally, we demonstrate zero-shot skill composition and lifelong sample efficiency. (iii) Reliable agents: We develop methods for agents to understand and execute both natural and formal language instructions, ensuring correctness and safety in task execution, particularly in real-world scenarios. By addressing these challenges, our framework represents a significant step towards achieving the FIRe desiderata in AI agents, thereby enhancing their utility and safety in a lifelong learning setting like the real world.
Evaluation of radiation damage on lutetium-aluminium and gold for practical applications using proton irradiation as a surrogate for neutrons
(University of the Witwatersrand, Johannesburg, 2024-10) Temaugee, Samuel Terungwa; Mavunda, Risimati D.; Usman, Iyabo T.
An understanding of the interaction of energetic radiations from particles such as protons, neutrons, and photons, with the microstructure of materials is crucial for predicting their bulk morphological response in extreme radiation environments. Exposure to these radiation species could lead to changes in the microstructural properties that, in turn, affect the mechanical and physical properties of the macroscopic matter. This thesis investigated the resilience of materials, specifically Au and Lu-Al, to radiation damage, employing computational simulation methods and experimental techniques. The study aims to provide critical insights into the radiation damage sturdiness of Au and Lu-Al, considering their application in reactor technology and other extreme radiation environments. Monte Carlo-based methods were employed to calculate radiation damage in the samples resulting from neutron and proton irradiation, utilizing MCNP6.2 and SRIM-2013, respectively. The objective was to compare ion beam irradiation with neutrons with a view to utilizing proton irradiation as a surrogate for neutron irradiation. Three different state-of-the-art characterization techniques—X-ray diffraction (XRD), High-Resolution Transmission Electron Microscopy (HRTEM), and Flash Differential Calorimetry(F-DSC)—were employed to evaluate damage in the materials before and after proton irradiation using the CLASS Accelerator at MIT, USA. The results of the study indicated that protons within the energy range 0.1 to 1.0 MeV produced similar types of damage in the materials as would neutrons (spectrum 0< E≤20 MeV at SAFARI reactor), suggesting protons as an alternative to neutron irradiation. Defect characterization in the materials using XRD and HRTEM techniques revealed dislocation loops and lines in both Lu-Al and Au, as well as Stacking Faults Tetrahedra (SFT) in the Au material. These defects with proton irradiation were similar to those observed with neutron irradiation in Au and other aluminum alloys. The estimated defect number density ranged from 0.7 to 4.8 × 1014 m−2, showing an increase with rising displacements per atom (dpa) or proton fluence post-irradiation. Lu-Al exhibited higher defect density values than Au, consistent with Monte Carlo simulations. Furthermore, results from the Flash DSC technique revealed significant changes in the characteristics of the power-temperature profiles (melting curves) of Lu-Al as dpa increased, offering insights into radiation-induced processes such as phase transition and precipitate stability at specific defect annealing temperatures. These findings are crucial for radiation damage studies for the binary alloy and warrant further investigation. The observed microstructural defect densities were relatively high, and prolonged exposure of the materials to higher doses could lead to further changes in microstructural properties, consequently influencing the physical and mechanical properties of the macroscopic material.
Physical property studies, tunnel numerical simulations and in-mine seismic experiments to image the gold orebody at South Deep Gold Mine
(University of the Witwatersrand, Johannesburg, 2024-09) Mulanduli, Omphulusa; Manzi, Musa
The investigation endeavors to assess the physical characteristics of deep borehole cores within the Upper Elsburg Reefs (UER) of the South Deep gold mine of the West Rand goldfield. Specifically, these cores are sourced from three boreholes situated approximately 2.6 km beneath the surface within the confines of the South Deep gold mine. The focal point of this study lies in non-destructive testing methods aimed at elucidating the intrinsic attributes of these rocks, with particular attention directed towards seismic velocities and densities. These measurements hold paramount importance in conducting numerical simulations to designing the in-mine (or tunnel) seismic reflection surveys acquired at South Deep gold mine, as part of the ERA-Min3 FUTURE (Fiber-optic sensing and UAV-platform techniques for innovative mineral exploration) project. Cultivating a profound comprehension of the seismic velocities and densities across diverse rock formations can significantly augment the interpretation of seismic reflections, thereby facilitating more refined assessments of subsurface geology and structural configurations. In pursuit of this goal, our study endeavors to delve into the fundamental acoustic properties of the gold-rich UER, with the overarching aim of deepening our understanding of its seismic reflectivity. To realize this objective, a comprehensive array of physical measurements, encompassing ultrasonic velocities and bulk densities, were conducted on drill-core specimens. To accurately portray the physical attributes of the lithological units under scrutiny, a total of twenty-four samples were subjected to exhaustive analysis for density and seismic velocity utilizing a spectrum of methodologies. Density determinations were procured through a diverse set of techniques, including dimensional assessments, employment of the KT20 MagSus tool, and utilization of the SNOWREX AHW-3 Professional Weighing Scale boasting a heightened sensitivity of 0.01 g. Ultrasonic measurements were undertaken employing the Proceq Pundit PL 2000 ultrasonic pulser velocity tester, equipped with two pairs of transducers boasting a center frequency of 54 kHz. The in-mine seismic survey was acquired to delineate geological structures that crosscut and displace the orebody. The study locale encompasses three distinct rock formations: the UER, gold-bearing conglomerate units (termed reefs), basaltic lava, and dyke specimens. The UER primarily comprises quartzites, exhibiting a P-wave velocity range of 5202-5802 m/s, an S-wave range of 3037-4768 m/s, and bulk densities spanning from 2.66 - 2.71 g/cm³. Conglomerate reefs exhibit a P-wave velocity range of 4467-5970 m/s, an S-wave range of 4040-4854 m/s, and bulk densities ranging from 2.67-2.94 g/cm³. Lava samples extracted from the boreholes showcase a P-wave velocity range of 5916 - 6711 m/s, an S-wave range of 3275-5659 m/s, and bulk densities spanning from 2.75-2.90 g/cm³. Singular dyke samples were encountered, exhibiting a P-wave velocity of 5921.5 m/s, an S-wave velocity of 5385 m/s, and a density of 2.85 g/cm³. The study employed the synth-seis code to simulate 1D seismic responses based on borehole data collected from the mine, aiming to validate findings from velocity and density measurements. Analysis of the seismograms indicated notable contrasts between conglomerates and quartzites, particularly evident in density and S-wave measurements, suggesting potential for improved rock discrimination with alternative seismic sources. Additionally, 2D numerical simulations were conducted to model wave propagation in the Upper Elsburg Reef (UER), revealing discrepancies between simulated and synthetic seismogram results, indicating potential limitations in seismic imaging. Furthermore, ray tracing was used to design a seismic survey inside the mine along the tunnel floor to image VCR (Ventersdorp Contact Reef) orebody and other geological structures. The real seismic survey was finally conducted inside the tunnel (SDT1), demonstrated the value of in-mine reflection seismic surveys for mapping geological structures at significant depths, which would otherwise be costly and logistically challenging. Despite noise interference from mine operations, processing algorithms enabled extraction of reflections and structural mapping from the dataset, underscoring the importance of such surveys in mining exploration and planning.
Applications of Recurrent Neural Networks in Modeling the COVID-19 Pandemic
(University of the Witwatersrand, Johannesburg, 2024-03) Hayashi, Kentaro; Mellado, Bruce
This study attempted to introduce moving averages and a feature selection method to the forecasting model, with the aim of improving the fluctuating values and unstable accuracy of the risk index developed by the University of Witwatersrand and iThemba LABS and used by the Gauteng Department of Health. It was confirmed that the introduction of moving averages improved the fluctuation of the values, with the seven-day moving average being the most effective. For feature selection, Correlation-based Feature Selection(CFS), the simplest of the filter methods with low computational complexity, was introduced as it is not possible to spend as much time as possible on daily operations due to providing information timely. The introduction of CFS was found to enable efficient feature selection.
Capability of multi-remote sensing satellite data in detecting and monitoring cyanobacteria and algal blooms in the Vaal dam, South Africa
(University of the Witwatersrand, Johannesburg, 2024-03) Obaid, Altayeb Adam Alsafi; Adam, Elhadi M.I.; Ali, Khalid A.
Vaal Dam is a large dam in South Africa. It is the primary source of potable water for the metropolitan and industrial areas of Gauteng province and other surrounding areas. The dam's surface area is about 320 km². It’s the second biggest dam in South Africa in terms of surface area, and it drains a catchment area of approximately 38,000 km². The dam's total capacity is about 2.603 × 10⁶ m³ (Haarhoff and Tempelhoff, 2007). The dam catchment area holds various anthropogenic activities, including major agricultural activities, mining, and some industrial activities (Obaid et al., 2023, Du Plessis, 2017), as well as many formal and informal settlements. The dam water is strongly affected by such activities, releasing chemical, physical, and biological contaminants and dissolved urban effluents, most of which enrich the nutrients that reach the dam water in some way. Water resources assessment and monitoring are crucial practices due to their direct contribution to the effective use of such resources. They require precise information about the water quantity and quality. Monitoring of inland water resources has been conducted using in-situ sampling and in-vitro measurement of the water quality constituents. However, these methods have limitations such as high cost, labor-intensive limited spatial and temporal coverage, and time consumption. Over the last few years, remote sensing has been examined for water quality monitoring as a cost- effective system. This research has tested satellite remote sensing to detect some water quality parameters in the Vaal Dam of South Africa. The main objective of this research is to examine the recent generation multispectral satellite sensors, Sentinel-2 MSI, and Landsat-8 OLI data to detect and assess chlorophyll-a and cyanobacteria in the Vaal Dam, South Africa to be used as a cost-effective monitoring tool. To achieve the objective, the research first aimed to understand how the spatial and temporal dynamics of land use, and land cover (LULC) impact algal growth in the dam reservoir. Land use land cover classification was conducted in the catchment area of the Vaal Dam using a pixel-based classification method. Landsat data for the period from 1986 to 2021 were classified using a random forest (RF) classifier in seven-year intervals (1986, 1993, 2000, 2007, 2014, and 2021). Applying the RF classifier revealed that overall classification accuracies (OA) ranged from 87% in the 2014 classified image to 95% in the 2007 image. The change-detection analysis revealed the continuous increase of the settlement class owing to the continuous population growth. A lot of anthropogenic activities associated with population growth have been recognized to release contaminants into the surrounding environment and might end up reaching the water resources causing significant deterioration. As a result, Vaal Dam encounters significant nutrient input from multiple sources within its catchment. This situation raised the frequency of the Harmful Algal Blooms (HABs) within the dam reservoir during recent years. The study also performed a time series analysis for the potential nutrients expected to be the enhancing factors for algal blooms in the Vaal Dam. Using chlorophyll−a (Chl−a) as a proxy of HABs, along with the concentrations of potential nutrients, statistical measures, and water quality data were applied to understand the trend of selected water quality parameters. These parameters were: Chl−a, total phosphorus (TP), nitrate and nitrite nitrogen NO₃NO₂_N), organic nitrogen (KJEL_N), ammonia nitrogen (NH₄_N), dissolved oxygen (DO) and the water temperature. The results reveal that the HAB productivity in the Vaal Dam is influenced by the levels of TP and KJEL_N, which exhibited a significant correlation with Chl−a concentrations. From the Long- term analysis of Chl−a and its driving factors, some very high values of Chl−a concentrations and its driving factors TP and KJEL_N were recorded in erratic individual dates which suggested some nutrients rich in wastes find their way to the dam. Another important notice was that the average Chl-a concentration significantly increased during the period of the study (1986 to 2023) it increased from 4.75 μg/L in the first decade (1990–2000) to 10.51 μg/L in the second decade (2000–2010) and reaching 16.7 μg/L in the last decade (2010–2020). Additionally, Chl−a data extracted from Landsat-8 satellite images was utilized to visualize the spatial distribution of HABs in the reservoir. The satellite data analysis during the last decade revealed that the spatial dynamics of HABs are influenced by the dam’s geometry and the levels of discharge from its two feeding rivers, with higher concentrations observed in meandering areas of the reservoir, and within zones of restricted water circulation. These spatial distribution patterns of HABs are associated with spatial variations of algal species in term of domination through the seasons of the year. The research also examined the utility of remote sensing techniques for mapping algal blooms using the current generation Sentinel-2 and Landsat-8 data. The effectiveness of some band ratio indices in the blue-green and red-near infrared wavelengths was tested. The results suggested that the blue-green band ratio of Landsat-8 [Rrs(560)/Rrs(443)], and red/NIR of Sentinel-2 [Rrs(705)/Rrs(665)] were found to be the best indices for Chl-a retrieval in the Vaal Dam. Results for the Landsat OLI dataset showed R² = 0.89; RMSE = 0.36 μg/L, P < 0.05, and the Sentinel MSI dataset revealed R² = 0.75; RMSE = 0.48 μg/L, P < 0.05 which is a high degree of accuracy. As the potential toxicity comes from the cyanobacterial bloom, the study examines different models to assess and map cyanobacteria concentration in the dam reservoir. Sentinel-2 and in-situ hyperspectral data have been used. None of the Sentinel-2 band ratios showed a significant correlation with the laboratory-measured values of the cyanobacteria. The in-situ measured Hyperspectra showed strong correlations between the band ratios Rrs(705)/Rrs(655) and Rrs(705)/Rrs(620), and the measured cyanobacteria (R² = 0.96 and R² = 0.95 respectively). Chlorophyll−a concentration was retrieved using band ratio indices in the red-NIR region. The strongest correlation was found between the retrieved Chl−a of band ratio Rrs(705)/Rrs(665) and the laboratory-measured Chl−a concentrations for both reflectance datasets. This correlation resulted in an R² value of 0.78 for Sentinel-2 reflectance data and an R² value of 0.93 for in-situ hyperspectral data. A Semi-analytical algorithm for estimating the Chl−a and phycocyanin (PC) pigments has also been examined. The algorithm uses the ratio of the calculated Chl−a absorption at 665 and phycocyanin absorption at 620 nm to their specific absorption coefficients a∗ (655) and a∗ (620) to estimate the concentration of Chl−a and phycocyanin respectively. It resulted in a strong correlation with measured chlorophyll-a, R² = 0.95. The algorithm also strongly correlated with measured cyanobacteria using the absorption to specific absorption ratio at 620 nm (R² = 0.97). However, the estimated values of cyanobacteria using a Semi-analytical algorithm resulted in cyanobacterial concentration values a little bit higher compared to the measured ones, hence, some factors used by the model need to be adjusted to the Vaal Dam site for better estimations. This research revealed that using band ratio indices of Landsat-8 and Sentinel-2 data are valuable tools for mapping chlorophyll-a in the Vaal Dam, a key indicator of phytoplankton biomass. Furthermore, using the semi-analytical algorithm with hyperspectral data is key for estimating the cyanobacteria concentration in the dam water. Models developed in this research will significantly improve near-real-time and long-term chlorophyll-a monitoring of the Vaal Dam. It will effectively help researchers and environmental agencies monitor changes in algal biomass of the dam water to address public health issues related to water quality. It helps to identify areas of high nutrient input and assess the effectiveness of water quality management strategies. It is of prime importance that the developments within the catchment of the Vaal Dam be carefully considered as it is one of the primary sources of dam water. The research recommends implementing the existing regulatory policies for effluent dispersal within the catchment to protect ecosystem functioning and water resources from further deterioration in their quality. It also recommends regular monitoring to detect real-time changes in HABs using satellite remote sensing.
Counting Reward Automata: Exploiting Structure in Reward Functions Expressible in Decidable Formal Languages
(University of the Witwatersrand, Johannesburg, 2024-07) Bester, Tristan; Rosman, Benjamin; James, Steven; Tasse, Geraud Nangue
In general, reinforcement learning agents are restricted from directly accessing the environment model. This restricts the agent’s access to the environmental dynamics and reward models, which are only accessible through repeated environmental interactions. As reinforcement learning is well suited for use in complex environments, which are challenging to model, the general assumption that the transition probabilities associated with the environment are unknown is justified. However, as agents cannot discern rewards directly from the environment, reward functions must be designed and implemented for both simulated and real-world environments. As a result, the assumption that the reward model must remain hidden from the agent is unnecessary and detrimental to learning. Previously, methods have been developed that utilise the structure of the reward function to enable more sample-efficient learning. These methods employ a finite state machine variant to facilitate reward specification in a manner that exposes the internal structure of the reward function. This approach is particularly effective when solving long-horizon tasks as it enables the use of counterfactual reasoning with off-policy learning which significantly improves sample efficiency. However, as these approaches are dependent on finite-state machines, they are only able to express a small number of reward functions. This severely limits the applicability of these approaches as they cannot model simple tasks such as “fetch a coffee for each person in the office” which involves counting – one of the numerous properties finite state machines cannot model. This work addresses the limited expressiveness of current state machine-based approaches to reward modelling. Specifically, we introduce a novel approach compatible with any reward function which can be expressed as a well-defined algorithm We present the counting reward automaton – an abstract machine capable of modelling reward functions expressible in any decidable formal language. Unlike previous approaches to state machine-based reward modelling, which are limited to the expression of tasks as regular languages, our framework allows for tasks described by decidable formal languages. It follows that our framework is an extremely general approach to reward modelling – compatible with any task specification expressible as a well-defined algorithm. This is a significant contribution as it greatly extends the class of problems which can benefit from the improved learning techniques facilitated by state machine-based reward modelling. We prove that an agent equipped with such an abstract machine is able to solve an extended set of tasks. We show that this increase in expressive power does not come at the cost of increased automaton complexity. This is followed by the introduction of several learning algorithms designed to increase sample efficiency through the exploitation of automaton structure. These algorithms are based on counterfactual reasoning with off-policy RL and use techniques from the fields of HRL and reward shaping. Finally, we evaluate our approach in several domains requiring long-horizon plans. Empirical results demonstrate that our method outperforms competing approaches in terms of automaton complexity, sample efficiency, and task completion.
Electrocatalytic detection of biomarkers of tuberculosis and cervical cancer
(University of the Witwatersrand, Johannesburg, 2024-07) Peteni, Siwaphiwe; Ozoemena, Kenneth Ikechukwu
The need for simpler, low cost and efficient diagnostic methods remains a matter of urgency. This has opened numerous streams of research. Electrochemistry is a simple, cost effective and efficient method that has been used for the detection of several diseases such as tuberculosis (TB) and human papilloma virus (HPV). TB has been ranked amongst the most problematic diseases in HIV/AIDS burdened communities, this alone calls for concern. Biomarkers of TB not only indicate mycobacterium infection but can also assist in the early detection of TB which is highly beneficial for the infected person and the health care system. HPV is the causative agent for cervical cancer. Cervical cancer is ranked as the fourth disease that causes mortality amongst women. With that in mind, HPV-16 L1 early detecting means possible early detection of cervical cancer. In this thesis, methyl nicotinate (MN), which is one of TB’s biomarkers was detected in phosphate buffer solution (PBS, pH 6.0) and commercial human serum using cobalt nanoparticles supported on carbon derived from trimesic acid (TMA) (abbreviated as Co-NPs@CTMA) and biphenyldicarboxylic acid (BPDC) abbreviated as Co-NPs@CBPDC) as electrocatalysts. These electrocatalysts were obtained using microwave-assisted metal-organic framework process with TMA and BPDC as ligands. XRD data showed that these electrocatalysts are cobalt nanoparticles with dominant {111} and {200} phase with traces of cobalt oxide (CoO). XPS and Raman data showed that Co-NPs@CBPDC is defect-rich compared to the Co-NPs@CTMA counterpart. BET showed that CoPs@CBPDC has higher surface area and pore size and volume than the Co-NPs@CTMA catalyst. Both electrocatalysts showed reversible cobalt nanoparticle oxidation and reduction reactions, in the absence and in the presence of the MN, thereby allowing for a facile indirect electrochemical detection of this biomarker. The calibration curves showed low limit of detection (LoD) of 0.47 and 0.147 µM for Co-NPs@CTMA and Co-NPs@CBPDC, respectively. The higher performance of the latter is attributed to its enhanced physico-chemical properties compared to the former. Next, HPV-16 L1, which is the conventional high-risk antigen that is present in cervical cancer, was detected using onion-like carbon (OLC) and polyacrylonitrile fibre integrated with OLC (OLC-PAN) as electrode platforms. Two electrode platforms were used; onion-like carbon (OLC) and its polyacrylonitrile (OLC-PAN) composites. Both platforms led to the detection in a wide linear concentration range (1.95 fg/ml to 50 µg/ml), excellent sensitivity (>5.2 µA/log([HPV-16 L1, fg/mL]) and ultra-low detection of ca. 1.0 and 1.4 fg/ml for OLC-PAN and OLC-based immunosensors, respectively. The high specificity of detection was proven by experimenting with an anti-Ovalbumin antibody (anti-Ova) and native Ovalbumin protein (Ova). An immobilized antigenic HPV-16-L1 peptide showed insignificant interaction with anti-OVA in contrast with the excellent interaction with anti-HPV-16 LI antibody. The immunosensors showed satisfactory stability of ~ 3 days of re-usability. The application of the immunosensor as a potential point-of-care diagnostic (PoC) device was investigated with the screen printed carbon electrode which showed the ability to detect ultra-low (~ 0.7 fg/ml) and high (~ 12 µg/ml) concentrations. This study opens the door of opportunity for further investigation with other electrode platforms and realization of PoC diagnostic devicesfor screening and testing of HPV biomarker for cervical cancer.
Exploring the Structure, Function and Stability of Glutathione Transferases Engineered from Intra- and Inter-class Consensus Sequences: How Forgiving is Nature?
(University of the Witwatersrand, Johannesburg, 2024-10) Mulenga, Thabelo; Achilonu, Ikechukwu; Sayed, Yasien
Protein folding is an enigmatic biochemical process that is foundational to the structural and functional requirements of a cell. The problem of protein folding, in a nutshell, concerns itself with the rate of protein folding as well as the conversion of amino acids from a linear sequence to a fully folded structure. This problem is partly answered by the existence of folding pathways. The folding funnel was conceptualised as a depiction of folding pathways, and it is a framework that illustrates that native proteins naturally favour the lowest energy state, encountering kinetic and thermodynamic barriers as they fold. Consensus protein design, based on this understanding, aims to: (1) enhance stability and (2) navigate the pitfalls of folding by modifying the folding funnel of a protein. This approach can also shed light on the significance of evolutionarily conserved residues. In this study, consensus protein mutants were generated for the Alpha and Mu glutathione transferases (GSTs) classes. The consensus proteins were then benchmarked against the parental proteins that were chosen (hGSTA1-1 and hGSTM1-1). The Alpha consensus mutant had 11 consensus mutations, including a notable M50L mutation, which affects the dynamic behaviour of helices α2 and α9, while the Mu consensus mutant had 13 unique mutations. Protein production and purification showed that the Mu consensus mutant had larger and purer yields. Data from far-UV circular dichroism studies and root-mean-squared-fluctuation (RMSF) from molecular dynamics (MD) simulations showed that the secondary structural components of the Alpha and Mu proteins remained largely the same, although the Alpha consensus mutant displayed a far lower molar residue ellipticity reading than its wildtype counterpart, indicating the disruption of secondary structural elements, likely caused by the M50L mutation. The ANS binding results showed that the M50L mutation in the Alpha consensus protein caused an increase in exposure of the surface area of the H-site, while the Mu consensus protein had a decrease in the solvent accessibility of its H-site. Thermal shift assay results indicated the consensus proteins had increased thermal stability. Enzyme kinetics results showed that the functionality of the proteins was severely diminished in the consensus mutants, particularly the Alpha consensus mutant. MD simulation results showed that there was an overall increase in the rigidity and compactness of the consensus mutant proteins, further affirming the improvement of thermal stability, while signalling the loss in functionality. The results produced herein have the potential to facilitate the proliferation of engineered GSTs for biotechnological applications that require proteins with an increased half-life and greater stability.
Flood Susceptibility Modeling in the uMhlatuzana River Catchment using Computer Vision-Based Deep Learning Techniques
(University of the Witwatersrand, Johannesburg, 2024-10) Chirindza, Jonas; Ajoodha, Ritesh; Knight, Jasper
In this study, covolutional neural networks (CNN) models are employed for flood susceptibility modeling in the uMhalatuzana River catchment in KwaZulu-Natal, South Africa. The CNN models, including 1D-CNN, 2D-CNN, and 3D-CNN, pro-vide a detailed assessment of flood vulnerability in the region. The models use di- verse spatial information, such as topography, land use, and hydrological features, to estimate the likelihood of flooding in different areas of the catchment. The flood susceptibility maps within the uMhalatuzana River catchment, classified into five risk zones namely, ‘very low’, ‘low’, ‘moderate’, ‘high’ and ‘very high’ susceptibility zone, serve as proactive instruments for risk mitigation and disaster management. The 1D-CNN model displays strong overall performance in flood susceptibility modeling, evident in key metrics such as accuracy, precision, recall, area under curve (AUC) score, and F1-score. The results suggest that the model effectively captures patterns in the input data, emphasizing its potential for flood susceptibility modeling. Moreover, the 2D-CNN model outperforms the 1D-CNN, achieving higher values when evaluated using various performance metrics. Finally, the 3D-CNN model outperformed both the 1D-CNN and 2D-CNN, emphasizing its predictive abilities in flood susceptibility modelling. The flood susceptibility maps produced by the 1D-CNN model, shows that most of the study area exhibits very low flood susceptibility (96.4%), with localized areas of higher susceptibility, particularly in the very high-risk category (2.53%). The 2D CNN model demonstrates a more diverse risk distribution, with a substantial portion having very low susceptibility (74.19%) and significant areas of higher risk, notably in the very high-risk category (10.93%). The 3D-CNN model emphasizes a spatial pattern where a large portion has very low susceptibility (84.10%), but with a concentration of high and very high-risk areas, comprising 12.34% of the total area. Finally, the consistent identification of higher risk susceptibility areas enhances the robustness of the assessments. The models’ high accuracy and detailed risk assessments provide valuable tools for decision-makers, urban planners, and emergency response teams in the uMhalatuzana River catchment. The precision of the models facilitates informed strategies for flood risk management, including targeted interventions such as improved drainage systems and early warning systems.
Envisioning the Future of Fashion: The Creation And Application Of Diverse Body Pose Datasets for Real-World Virtual Try-On
(University of the Witwatersrand, Johannesburg, 2024-08) Molefe, Molefe Reabetsoe-Phenyo; Klein, Richard
Fashion presents an opportunity for research methods to unite machine learning concepts with e-commerce to meet the growing demands of consumers. A recent development in intelligent fashion research envisions how individuals might appear in different clothes based on their selection, a process known as “virtual try-on”. Our research introduces a novel dataset that ensures multi-view consistency, facilitating the effective warping and synthesis of clothing onto individuals from any given perspective or pose. This addresses a significant shortfall in existing datasets, which struggle to recognise various views, thus limiting the versatility of virtual try-on. By fine-tuning state-of-the-art architectures on our dataset, we expand the utility of virtual try-on, making them more adaptable and robust across a diverse range of scenarios. A noteworthy additional advantage of our dataset is its capacity to facilitate 3D scene reconstruction. This capability arises from utilising a sparse collection of images captured from multiple angles, which, while primarily aimed at enriching 2D virtual try-on, inadvertently supports the simulation of 3D environments. This enhancement not only broadens the practical applications of virtual try-on in the real-world but also advances the field by demonstrating a novel application of deep learning within the fashion industry, enabling more realistic and comprehensive virtual try-on experiences. Therefore, our work heralds a novel dataset and approach for virtually synthesising clothing in an accessible way for real-world scenarios.

Faculty of Science (ETDs)

Browse

Filters

Settings

Sort By

Results per page

Search Results