42337 research outputs found
Sort by
Conflict and conservation: evaluating land cover dynamics in the tri-border region of Rwanda, Uganda, and the DR Congo
No full textThe environmental impact of armed conflict is often an under-considered but important aspect of post-conflict recovery. The variety and complexity of these impacts can be extensive, typically varying with the conflict’s intensity, scale, and duration. Land cover is often used as a proxy to explore environmental change, but diverse and insecure environments are generally not captured sufficiently on global land cover maps, necessitating development of site-specific maps. This research investigated the impact of different types of conflict around four protected areas in a tri-border area between Rwanda, Uganda, and the Democratic Republic of the Congo (DRC). Land cover maps were developed using Random Forest in Google Earth Engine for four time periods from 1987 to 2024 to measure change over time. Findings indicate that areas near conflict zones in Rwanda and the DRC underwent more significant land cover changes compared to Uganda, which experienced greater changes farther away from conflict. Notably, the DRC displayed the most dynamic range of land cover shifts across all time spans, possibly mirroring its similarly wide variation in conflict type and length. The national parks exhibited minimal land cover change during and after conflict. This resilience is likely attributed to substantial conservation investment and natural barriers imposed by challenging topography. This research underscores the nuanced relationship between armed conflict and environmental change, highlighting the need for targeted environmental strategies in post-conflict settings
Investigating coastal dune erosion dynamics across Lundin Links golf course using high-resolution UAV imagery
No full textCoastal dune systems are dynamic environments that provide vital natural protection against storm surges and sea level rise, yet many are experiencing accelerated erosion under growing climatic and anthropogenic pressures. This study assesses recent morphological changes in a small coastal dune system at Lundin Links, Scotland, using high resolution UAV LiDAR and UAV SfM photogrammetry datasets. The survey employed RTK-enabled drone platforms and generated precise terrain models supported by ground control points. A comparative analysis of datasets reveals strong geometric consistency between two datasets. Given its lower cost and simplicity, photogrammetry proves sufficient for capturing small scale dune dynamics in open coastal environments. Multitemporal digital elevation models derived from UAV and historical LiDAR data, volumetric and spatial analyses show a shift from initial localised accretion to widespread erosion, with an acceleration in sediment loss after 2020. Change detection highlights crest lowering and landward retreat, raising concerns about future exposure in areas adjacent to golf course fairways. These changes align with national coastal change trends and is likely exacerbated by recent extreme weather events. This research demonstrates the effectiveness of UAV SfM and support more frequent, accessible monitoring approaches to inform adaptive management under increasing climatic risk
Woodland creation: monitoring natural processes and hybrid approaches
Get PDFCreating new woodlands via natural processes or hybrid methods that combine planting and natural processes can be a great way to support nature recovery and reduce resource use (e.g. reduced need for nursery tree production, tree protection etc). However, the process is highly variable and keeping an eye on how things are going will help to ensure the woodland is developing as you intended. The method described here allows land managers, volunteers or any other interested parties to survey sites to collect robust data on the progress of natural processes regardless of experience level. The method is flexible to fit the surveyors’ objectives and allows for comparisons over time providing it is carried out the same way at each time point. This method was developed by academics working on the TreE PlaNat project and reviewed the the Knowledge User Board
Analysis of Vi antigen expression and function in Salmonella enterica serovar Dublin
Get PDFThe Vi antigen is a capsular polysaccharide commonly associated with Salmonella enterica serovar Typhi. It contributes to bacterial evasion of innate immune responses in vitro, however its role during infection is less well understood as S. Typhi is an obligate human pathogen. The Vi antigen is occasionally detected in Salmonella enterica serovar Dublin, which naturally causes typhoid-like disease in cattle, thereby offering a model to dissect its contribution to pathogenesis. Whole-genome sequences were obtained for three strains of S. Dublin reported to be Vi-positive. High molecular weight genomic DNA was extracted and long-read sequencing was performed using the Oxford Nanopore PromethION platform. Short-read sequencing was also performed with the same DNA on the Illumina NextSeq platform. Hybrid genome assemblies were produced by constructing a long-read assembly and polishing with short-reads. Analysis showed that the viaB locus encoding the Vi antigen and wider Salmonella Pathogenicity Island 7 region were largely conserved in the three strains. However, variation in genome structure, sequence types and plasmid repertoire were detected between the strains. While one strain did have the viaB locus, no Vi polysaccharide could be detected by agglutination or blotting with Vi-specific antiserum. This may be associated with a deletion detected in the tviB gene of this strain. To define the role of viaB-encoded genes, deletion mutants were created that lack the viaB locus, a viaB-encoded transcriptional regulator (tviA), and genes associated with polysaccharide export (vexA, vexBC, vexD and vexE). These were shown by whole-genome sequencing to lack off-target mutations. Loss of tviA reduced but did not abolish surface expression of Vi antigen, whereas all other mutants failed to produce the antigen. The mutants will be a valuable resource to study the role of the Vi antigen in interactions between S. Dublin and its bovine host, including in oral challenge and surgical models of infection to measure invasion, inflammation and systemic translocation
The theory and application of non-stationary and deep Gaussian processes in regression problems
No full textThe focus of this work is the convergence of non-stationary and deep Gaussian process
regression. More precisely, we follow a Bayesian approach to regression or interpolation,
where the prior placed on the unknown function f is a non-stationary or deep Gaussian
process, and we derive convergence rates of the posterior mean to the true function f in
terms of the number of observed training points. In some cases, we also show convergence
of the posterior variance to zero. The only assumption imposed on the function f is that
it is an element of a certain reproducing kernel Hilbert space, which we in particular cases
show to be norm-equivalent to a Sobolev space. Our analysis includes the case of estimated
hyper-parameters in the covariance kernels employed, both in an empirical Bayes setting and
the particular hierarchical setting constructed through deep Gaussian processes. We consider
the settings of noise-free or noisy observations on deterministic or random training points.
We establish general assumptions sufficient for the convergence of deep Gaussian process
regression, along with explicit examples demonstrating the fulfilment of these assumptions.
Specifically, our examples require that the Hölder or Sobolev norms of the penultimate layer
are bounded almost surely.
In addition to these theoretical results, we present numerical simulations that further validate
our findings. For instance, we demonstrate that even with a ’bad choice’ of a non-stationary
kernel- where the prior does not match the non-stationary structure of the unknown function
f- the theoretical convergence results remain robust. Moreover, the simulations reveal
that certain theoretical assumptions for deep Gaussian processes, such as the need to
reduce regularity hyperparameters in ascending layers of the deep Gaussian process, are
often unnecessary in practice. We apply non-stationary and deep Gaussian processes to a
variety of synthetic and real-world datasets, both to illustrate the theoretical insights and
to benchmark the performance of our methods
Disciplining bestial hunger
Get PDFThe thesis observes the motif of bestial hunger and the disciplinary mechanisms surrounding it, either intended to control it or to further enable it. In an effort to identify more nuanced attitudes towards the way hunger functions three different texts have been selected.
Firstly, the chivalric romance, The Tale of Sir Gowther, which features the motif of hunger in the contexts of childhood, human and animal boundaries, penitence, noble and chivalric conduct, and sainthood. The narrative displays a multitude of literary and cultural influences, while being a psychoanalytically rich exploration of the individual’s process of socialisation.
Secondly, the thesis discusses the compendious dream vision, The Visions of Piers the Plowman, which features several narratives which the thesis focuses on. The Rat Fable, a shorter narrative in the text’s Prologue is observed from the perspective of political sophistry giving rise to unrestrained consumption. The second part focuses on the episode featuring Piers’s interactions with Hunger embodied on the Half-Acre, and the further resonances of
the encounter in the narrative.
Finally, Robert Henryson’s Moral Fables are observed, where the classical Aesopic animal fables provide a framework for the poet’s exploration of worldly injustice. This chapter follows different characters, among some others the Fox and the Wolf throughout the fable
collection, observing the ways they avoid and use language in order to satiate their own voracity.
The thesis ultimately aims to observe the different manifestations of hunger in these late medieval texts. It synthesises multidisciplinary (psychoanalytical, ecological, economical) approaches in an effort to observe the ways the motif influences the characters and the narratives
Investigating London dispersion forces in solution
No full textChapter 1 focuses on weak interactions which are crucial from macroscopic to atomic levels, playing a vital role in processes such as the folding and stabilization of biomolecules to chemical selectivity and various other contexts. One component of weak interactions is called van der Waals forces, which includes Keesom, Debye, and London dispersion forces. London dispersion forces, although the weakest in terms of energy, are the predominant type of interaction among apolar moieties like alkyl groups. Molecular balances have historically been the preferred method for studying weak interactions like hydrogen bonds. However, studies focusing on London dispersion forces using molecular balances often yield drastically variable results, influenced by several parameters unique to each investigation.
Despite being an attractive component of all weak interactions, London dispersion (LD) is often overshadowed by stronger forces. Chapter 2 presents the synthesis of terphenyl molecular torsion balance variants to investigate the thermodynamic aspect of LD forces between alkyl chains. NMR techniques were employed to measure this parameter, revealing the complex interplay of factors influencing the conformational equilibrium of the balances across different solvents. The LD component was found to be weak, and sometimes outcompeted by weak H-bonding interactions. While solvent polarity affected the equilibrium, factors such as solvophobicity (as encoded by cohesive energy density) and change in solvent accessible area (SASA) proved to be important in accounting for the observed trends. Computational studies suggested that a weak intramolecular hydrogen bond could occur between the protons alpha to the carbonyl group on one chain and the ester carbonyl oxygen on the other was identified. This hypothesis was supported by experimental results from the EtOMe balance variant.
Chapter 3 presents the investigation of the kinetic aspect of LD forces between alkyl chains. The molecular torsion balance used in the second chapter and various NMR techniques were employed to measure the barriers to rotation. The effects of the solvent and identity of the appended alkyl chains on the kinetics of conformer exchange was investigated using EXSY NMR spectroscopy. Polar solvents
consistently exhibited lower rotational barriers compared to apolar solvents, underscoring the significant role of intermolecular hydrogen bonding in stabilizing the transition states of the terphenyl balance. The trend of decreasing rotational barriers with increasing chain length in three different apolar solvents indicates that LD forces stabilize the transition state. Additionally, the findings in carbon disulfide highlight the contribution of bulk polarizability to intermediate stabilization, particularly for the balances containing larger alkyl groups. Overall, the results advance the understanding of solvent interactions (mediated by both electrostatics and LD) in stabilizing rotational transition states.
Chapter 4 presents a comparison of the kinetics of conformational interconversion in molecular torsion balances hosting alkyl-alkyl, chalcogen bonding, and hydrogen bonding interactions. 1D EXSY NMR was utilized to quantify the solvent dependence of breaking and forming these classes of interactions. The conformational dynamics of balances hosting alkyl-alkyl interactions and chalcogen bonds were much less influenced by solvent polarity than those containing hydrogen bonds. This reflects the increased polarity and longer electrostatic range of H-bonds compared to other interactions.
One counterintuitive finding was that the kinetics of conformational exchange in the H-bonding balances was increased by polar solvents. In apolar solvents the intramolecular H-bond was formed in the folded conformation and also stabilized the transition state via long-range electrostatic interactions. In polar solvents this energetic benefit was lost because solvation of all states was increased. However, stronger solvation by H-bonding was suggested to introduce deeper energetic troughs into the conformational landscape by increasing energy differences between the ground and transition states. Accordingly, greater solvent- and system-dependent behaviour was observed for the more polar balances. Moreover, since energy barriers to conformational exchange are determined by the difference between the ground states and the transition state, it can be challenging to determine the precise origins of the observed energetic effects without very careful consideration of the structure of the system and associated solvation states. While carbon disulfide and benzene yielded the lowest kinetic barriers to conformational exchange for the formyl balances, solvent-mediated London dispersion does not appear to provide a general lubricating influence on the kinetics of exchange, but arises instead from their apolar, weakly interacting nature.
The comparative data revealed that the alkyl-containing terphenyl balances presented much smaller solvent dependencies on their conformational kinetics coupled with smaller energy differences between the folded and unfolded states. The data point towards mankind’s selection of long-chain alkanes as lubricants as arising from the smoothing of conformational energy landscapes. Apolar chains are weakly interacting and the London dispersion interactions that they participate in are non-directional, which flattens the energy landscape between conformational states.
The findings in this thesis may enhance the accuracy of existing computational methods, providing more precise models for simulating molecular interactions. Additionally, this research could contribute to a deeper understanding of the molecular origins of phenomena such as friction and viscosity, potentially leading to advancements in the design and application of lubricants and other materials
Free trade agreements through a prism: exploring the making of trade policies from a subcentral government context
No full textFree Trade Agreements (FTA) have become more complicated, complex, and progressive with deeper policy integration over time (Capling and Low 2010b; WTO 2011). Where early FTAs addressed at-the-border trade barriers, such as tariffs, contemporary agreements include non-tariff barriers and social policy obligations. Since the 1990s, FTAs started to include non-tariff barriers including labour, environment, transparency, and public procurement measures. These ‘deep’ obligations are found in most modern FTAs. The impact of the growing complexity and remit of FTAs is that they now include policy matters which are managed by subcentral governments and where they have ultimate jurisdiction on these policies, including in Canada and the UK (Melo Araujo 2017a; Kukucha 2015; 2003; Paquin 2013). As a result, these subcentral jurisdictions have been pulled onto the international stage because they have policy jurisdiction over many FTA obligations.
Without formal policy levers at their avail, neither a seat at the negotiation table nor a pen to sign with, subcentral jurisdictions are shaping and influencing the outcomes of FTAs, albeit with varying degrees of success. How this is done and which actors are involved in these FTA policy-making processes is not well-known. My research explores this gap by following the making of the Canada-UK FTA negotiation process from the subcentral government context – with the Canadian provincial and UK devolved governments. To do this, I ask: From the subcentral government context, how are FTAs made and in what ways are subcentral government interlocutors involved in these processes?
Theoretically, this research is positioned within the post-human new materialist paradigm leaning on Karen Barad’s (2007) Agential Realism and Bruno Latour’s (2005) Actor-Network Theory as the basis for the development of my Prism Approach. This research brings forward a discussion about the material complexities and nuances involved in the creation of these complex legal promises which occur in conjunction with additional becomings (e.g., trade negotiators, teams, and networks) through these cocreated processes.
This research incorporated a multi-phased qualitative research design bringing together methods which organically fit with the research question and philosophical underpinnings. The five phases included: policy literature review, interviews with subcentral government policy professionals, series interviews with subcentral government policy professional key informants, participant observation case study of a devolved government trade policy team incorporating a social network analysis component, and two roundtable workshops bringing together the devolved and provincial government trade policy delegations.
This research offers contributions to theoretical discussions building on Barad and Latour’s scholarship by reading them diffractively to generate my Prism Approach, and empirically by providing details of the making of an FTA highlighting the reasons why subcentral governments are involved, the processes involved in these negotiations, and an ethnographic case study of a devolved government trade policy team’s network of interlocutors
Machine learning in drug discovery: advancing protein-ligand binding affinity predictions
No full textBinding affinity quantifies the strength of the interaction between a protein and
a small drug-like molecule. Accurately determining binding affinity helps identify
promising drug candidates in the early stage of drug discovery, particularly in hit discovery
and lead optimization phases, where screening several millions to even billions
of compounds is required. Hit discovery involves identifying potential compounds
(known as ‘hits’) that show initial activity against the choice of disease-causing protein
target. Lead optimization focuses on refining these hits to improve their binding
affinity and other drug-like properties. Experimental assays are the gold standard
for determining binding affinity, but they are not practical for rapidly screening
millions of drug-like compounds against potential targets. Accurate in silico prediction
of protein-ligand binding affinity can significantly expedite drug discovery by
streamlining the identification and optimization of viable drug candidates, reducing
huge experimental costs and time.
Over the last fifty years, a wide range of in silico binding affinity prediction
strategies have been developed. They consist of both structure-based and ligand-based
approaches. However, these methodologies often fall short in large-scale
screenings. So called docking methods, while capable of high-throughput screenings,
often lack the desired accuracy for a binding affinity prediction. In contrast,
alchemical free energy-based (AFE) techniques, a simulation-based technique, offer
improved accuracy but are computationally demanding. The rapid progression of
machine learning, coupled with increased accessibility to binding affinity data, opens
avenues to deep learning-based methods for improving the accuracy and speed of
binding affinity predictions.
This thesis focuses on exploring and developing machine learning methods for
predicting protein-ligand binding affinity. The first part of the thesis investigates
how current deep learning models learn from input protein and ligand data to predict
binding affinity. Systematic experiments using publicly available kinase datasets are
conducted to assess the impact of protein encodings and ligand encodings derived
from convolutional and/or graph neural networks by inputting variations of protein
and ligand data. The results indicated that protein encodings have minimal impact
on binding predictions, while ligand-based features play a more substantial role in
model performance.
The second part of the thesis focuses on addressing key challenges at the model,
data, and evaluation levels of the deep learning framework for predicting binding
affinity. To overcome challenges at the model level, this work introduces a deep learning
framework for predicting binding affinity using pretrained protein and ligand
language models, called BALM. Utilizing pretrained language models for proteins
and ligands, the BALM method predicts binding affinity by optimizing the distance
between protein and ligand encodings using a cosine similarity metric. At the
data and evaluation levels, the research demonstrates novel strategies for training
and testing these models to ensure they provide meaningful and reliable predictions
compared to traditional methods and experimental measurements. While zero-shot
prediction on unseen targets may not always be reliable, the few-shot finetuning of
the BALM model is shown to be reliable for screening new targets, demonstrating
better performance than docking.
The final part of this thesis focuses on integrating machine learning with physics-based
simulation methods, such as alchemical free energy calculations. This integration
aims to reduce computational costs and time during lead optimization.
Specifically, Active Learning (AL) is used to intelligently select compounds for AFE
calculations, making the identification of top binders more efficient. AL is an iterative process that learns binding affinities from an unlabelled dataset and helps
prioritize compounds for detailed evaluation, minimizing the need to compute AFE
for all compounds in a large pool. In this approach, machine learning models such
as Gaussian process regression and pretrained graph neural network-based models
act as surrogate models during each AL iteration. They provide predictions of
binding affinity to inform the next set of AFE calculations, allowing for efficient
compound selection. Various recommendations on model choice, batch size, and
strategies for exploring or exploiting the chemical spaces based on the ligand pool
used are provided. Both models show similar recall in identifying top binders on
large datasets, but the Gaussian process model performs better than the pretrained
graph network model when the training data is limited. Using a larger initial batch
size, especially with diverse datasets, improved recall for both models and enhanced
overall correlation metrics. However, smaller batch sizes were more effective for later
iterations
Enhancing trust in NIDS: from flawed benchmarks to formal guarantees
Get PDFNetworks serve as the critical backbone for our increasingly digital world, supporting everything from personal communications to essential infrastructure. Securing these networks against malicious activities is paramount for maintaining the confidentiality, integrity, and availability of data. Machine learning (ML) has been extensively applied to network intrusion detection, with researchers particularly interested in ML models' ability to detect generalise to new attack patterns. However, research into Network Intrusion Detection Systems (NIDS) can be marred by their opaque nature. These models often lack transparency on two fronts: they are trained on large datasets with flaws that compromise benchmarking validity, and they function as black boxes, making critical security decisions via mechanisms that resist straightforward interpretation. These issues fundamentally undermine trust in their capabilities. This thesis addresses this trust gap via two complementary strands of research: the systematic interrogation of benchmark NIDS datasets and the application of neural network verification techniques to NIDS.
We begin by introducing the concept of 'Bad Data Design Smells' as indicators of flaws in the design of synthetic datasets that undermine their suitability as evaluation benchmarks. Through a systematic literature overview and detailed case studies, we demonstrate how these flaws significantly impact downstream research, such as dataset artefacts degrading classification accuracy by over 90%. We develop a two-pronged approach to identify these smells, combining systematised manual analysis with automated heuristic measurements. We then extend this analysis by contextualising the complexity of network data. We introduce a novel metric based on spectral clustering that allows us to compare NIDS benchmarks with datasets from other fields. Despite their ubiquity, our measurements consistently reveal that benchmark NIDS datasets exhibit minimal complexity compared to even simple benchmarks in other domains, limiting their utility in research.
Building upon these insights, we use neural network verification techniques to improve our ability to reason about ML-based NIDS training and inference. First, we encode domain knowledge as formal specifications that models must adhere to, enforcing these properties through adversarial training. We then formally verify model adherence, which substantially improves cross-dataset generalisation accuracy by over 35%. In addition, we use these specifications as a form of explainability, gaining insight into model decision boundaries and failure modes, such as ranking model fragility to feature perturbations. Our approach outperforms model certification techniques, which we show fails in likely settings. Second, we leverage the counter-examples produced by verification frameworks to generate constrained, realisable adversarial examples for NIDS, addressing a notable gap in the field between feature-space and problem-space adversarial attacks and improving on standard adversarial attacks in some cases.
By adopting a strategy of both scrutinising dataset quality and formally defining model behaviour, this thesis improves the trustworthiness of NIDS models. We address trust at both training and inference, by establishing clear links between dataset flaws and model outcomes, as well as formally specifying model behaviour to resist evasive attacks and concept drift. Altogether, these contributions introduce greater rigour into NIDS evaluation processes, advancing both theoretical and practical aspects of ML applied to network security
Access Repository Dashboard
Do you manage Edinburgh Research Archive? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!