Our PhD projects

Doctoral researcher: Melina Müller
Supervisory team: Nina Kudryashova, Matthias Hennig, Luke Bashford
Project partner: Newcastle Hospitals NHS Foundation Trust
Abstract: By stimulating the brain in the correct way, patients with neural disorders can be helped. For instance, stimulations can be used to to put sensory information about an external device directly into a person’s brain. This is useful for brain computer interfaces (BCI), which are external devices that are controlled by a person’s brain activity. To accomplish this, the influence stimulations have on the brain’s activity needs to be understood. Once that is known, the right stimulation needed to archive a specific brain state can be chosen, i.e. brain activity can be controlled. However this relationship between stimulations and brain activity is very complex since they can have strongly varying effects. The aim of this thesis is to develop a model that can control brain activity by learning from past stimulation effects and considering the current activity. For this, a method will be developed that carries out the most informative stimulations for data collection. This data will then be used to quantify stimulation effects on brain activity and based on that how to choose the stimulation that achieves a target activity. These methods will be developed such that they are complex enough for real-life settings like brain computer interfaces.

Doctoral researcher: Bianca Branco
Supervisory team: Alex Kwong, Heather Whalley, Peggy Series
Project partner: Mental Health Platform
Abstract: Depression is a complex and heterogeneous disorder that affects millions of people worldwide. Despite advances in psychiatric research, important challenges remain in understanding why individuals differ in their susceptibility to depression, how symptoms fluctuate over time, and why clinical outcomes vary so substantially across patients. Addressing this heterogeneity is a key challenge for precision psychiatry, as current diagnostic categories often fail to capture the diversity of underlying biological, environmental and clinical factors associated with depression. The increasing availability of large-scale population cohorts, longitudinal healthcare records, genetic data and digital health technologies presents new opportunities for studying depression through a data-driven lens. Machine learning methods are particularly well suited to integrating these diverse sources of information, enabling the identification of predictive factors, latent population structure and temporal patterns that may not be apparent using traditional analytical approaches. This PhD investigates depression heterogeneity across multiple timescales using supervised and unsupervised machine learning techniques. Leveraging data from large population cohorts and ecological momentary assessment studies, the research aims to characterise variation in depression risk, symptom dynamics and long-term trajectories. Particular emphasis is placed on understanding how genetic and environmental factors contribute to individual differences in depression, and whether these differences can be used to identify more informative and clinically relevant patterns of disease progression. By combining diverse data modalities and longitudinal perspectives, this work aims to advance our understanding of depression heterogeneity and contribute to the development of more personalised approaches to prevention, diagnosis and treatment.

Doctoral researcher: Jamie Davies
Supervisory team: Mehrdad Yaghoobi, Lucy Kershaw
Abstract: Cardiac cine provides videos of the beating heart and is central to the clinical assessment of cardiac function. Its core limitation is a trade-off between spatial resolution, temporal resolution, and signal-to-noise ratio. Resolving cardiac motion requires rapid data acquisition, but undersampling reduces spatial resolution, introduces aliasing artefacts, and lowers signal-to-noise ratio. Classical reconstruction supports only moderate undersampling, so routine clinical practice relies on ECG-gated segmented acquisition, which combines data from multiple heartbeats into each frame. This assumes beat-to-beat consistency and requires breath-holding, excluding patients with arrhythmia or breathing difficulties, those for whom cardiac cine is most diagnostically important. Real-time methods do exist as a fallback, but compromise on image quality and reconstruction speed. Deep learning algorithms address these challenges, but topperforming methods are supervised, requiring fully sampled references for training. Such references are fundamentally compromised. They are either limited to compliant patient cohorts or capped at the quality of the classical reconstructions used to build them, and in both cases cannot represent realtime dynamics. This project investigates self-supervised methods that train directly from undersampled data, with the goal of clinically robust real-time cine imaging, under either a single breath-hold or free breathing. Progress includes establishing classical and supervised baselines on CMRxRecon2023, and evaluating several self-supervised losses for multi-coil cardiac cine against the supervised baseline.

Doctoral researcher: Artur Miralles Méharon
Supervisory team: Chris Ponting, Sjoerd Beentjes, Ava Khamseh
Project partner: PrecisionLife
Abstract: Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a debilitating, multisystem disease which is widespread but currently has no reliable cure, treatment or diagnostic test. There is wide variation in symptom profiles, severities and comorbidities, posing a challenge for biomarker discovery and classification. Growing evidence suggests that ME/CFS is not a single homogeneous condition but instead covers a spectrum of disease processes, potentially driven by diverse biological mechanisms. This year’s work consists of two parts. The central focus of part 1 has been to investigate heterogeneity in ME/CFS through stratification using blood trait data from the UK Biobank. Various methods for stratification have been considered: a hierarchical stochastic block model, hierarchical clustering with selective inference and Stator. In part 2, I conduct complementary analyses of incidence and comorbidity patterns of ME/CFS in the UKB and All of Us Research Program (AoU) cohorts with the aim of replicating and extending previous findings. Collectively these approaches aim to contribute to a more refined characterisation of ME/CFS.

Doctoral researcher: Rodrigo Lara Molina
Supervisory team: Michael Gutmann, Andrea Weisse
Project partner: Danu Insights
Abstract: Antibiotic resistance is a critical global health challenge, threatening the effectiveness of modern medicine. Understanding the mechanisms behind resistance is essential for developing strategies to combat it. This project focuses on designing optimal experiments for mechanistic models of RNA repair-mediated antibiotic resistance in E. coli. Specifically, we study the Rtc system, a gene network for RNA repair, which provides transient heteroresistance to antibiotics. Using Bayesian Experimental Design (BED), we aim to identify experiments that maximize the information gained about key model parameters and that distinguish between competing hypotheses of the Rtc system’s role in resistance. Our approach leverages Machine Learning (ML) to approximate intractable computations within the Bayesian framework, enabling efficient optimization of experimental designs. Specifically, in this report we focus on the preceding task of prior elicitation using simulator-based inference. The ultimate goal is to produce and validate experimental designs which are informative, and this notion depends heavily on our prior. By advancing the application of BED to molecular systems biology, this work aims to contribute to the understanding of RNA repair mechanisms, and to offer insights for improving ribosome-targeting antibiotics, which constitute a significant portion of prescribed antibiotics globally.

Doctoral researcher: Núria Fàbrega
Supervisory team: Ian Simpson, Kenny Baillie, Mirella Lapata
Abstract: Some of the evidence needed for new scientific discoveries may already be sitting in public research databases, waiting to be reused. Data from previous biological studies can help answer new questions, strengthen findings by comparing or combining evidence across studies, and reveal where new research is still needed. Yet much of this potential is currently lost. Important details about how studies were carried out are often incomplete, or scattered across databases and scientific papers, meaning that checking whether even one study is suitable for a particular question can take considerable manual work. My PhD aims to make one type of study easier to reuse: transcriptomics studies, which measure how active different genes are in different groups of people, tissues, or conditions. I am developing a system that automatically reconstructs what these studies did and identifies those that fit what a researcher is looking for. It combines AI with standardised biological vocabularies to recognise when studies describe the same thing in different ways. The system will also show where its information came from and how confident it is, so that researchers can check its decisions rather than simply trust an automated answer. After identifying suitable studies, the system will compare their biological results. This will help researchers see how consistent the findings are across studies, understand where they differ, and identify cases where further research may be needed. By making it easier to identify useful existing studies, this work aims to help the research community get more value from the data that have already been collected, supporting faster scientific progress and more reliable findings.

Doctoral researcher: Binjie Chen
Supervisory team: Giovanni Stracquadanio, Eve Miller-Hodges
Abstract: Lysosomal storage disorders are rare genetic diseases in which people cannot make crucial lysosomal enzymes. These are importantproteins that help break down certain substances. Without these enzymes, harmful materials can build up in the body and damage organs over time. Treatments called enzyme replacement therapy (ERT) exist for some of these conditions, but are very expensive, hard to produce, and often not very effective. Mu_x0002_copolysaccharidosis Type IV A (MPS IVA), is a lysosomal storage disorder which causes severe problems with the airways, heart valves, bones and joints. An ERT treatment exists for MPS IVA, but is not very effective. This project aims to use artificial intelligence (AI) to design better, more affordable enzymes for treating MPS IVA. By training computer models on large databases of protein structures and sequences, we can create new versions of the missing enzyme that work better in the human body. Promising designs will then be tested in lab-grown cells to check their safety and function. Our goal is to make future treatments more effective and accessible for patients living with rare enzyme-related diseases. This work also helps pave the way for using AI more widely in drug development, especially for conditions that currently have limited treatment options.

Doctoral researcher: Rishi Ramessur
Supervisory team: Steven McDonagh, Sotirios Tsaftaris, Luciana D'Adderio, Peter Thomas
Project partner: Moorfields Eye Hospital
Abstract: Multimodal AI systems have considerable potential to support healthcare delivery. In Ophthalmology, the largest UK outpatient specialty with over 10 million annual appointments, these systems could help address rising demand and workforce pressures by supporting referral assessment, information extraction, clinical audit, and other administrative or decision-support tasks. However, despite strong performance on benchmark datasets and simulated scenarios, current VisionLanguage Models (VLMs) remain unreliable in real-world healthcare settings, where they may be poorly calibrated, overconfident, vulnerable to distribution shift, and unable to reliably communicate uncertainty. Our preliminary evaluation of VLMs on real-world ophthalmology referrals demonstrated extraction accuracies as low as 36.3%, despite substantially stronger performance on synthetic or idealised inputs, highlighting an important translational gap between benchmark performance and safe clinical deployment. This project will develop a clinically grounded framework for evaluating uncertainty and robustness in multimodal ophthalmic AI systems. The work will focus on how combinations of medical images and written clinical information influence prediction confidence, calibration, abstention behaviour, and human decision-making. The project aims to move beyond predictive performance alone by understanding when AI systems become unreliable, when they should appropriately defer to clinicians, and how uncertainty can be communicated safely and meaningfully in clinical practice.

Doctoral researcher: Emilia Agasi
Supervisory team: Ian Simpson, Charlie Gourley
Abstract: High-grade serous ovarian carcinoma (HGSOC) is characterised by significant inter-patient heterogeneity and high mortality rates, making accurate prognosis a persistent challenge. Patient similarity networks (PSNs) offer a way to model relational structure between patients. However, their effectiveness depends on the quality of learned representations. In particular, representations derived from generic pre-trained models often fail to capture task-relevant structure and commonly used similarity metrics often do not reflect cross-modal relationships. Methods: We propose a representation-driven framework for multimodal graphs-based survival prediction that integrates histopathology, transcriptomics and clinical data from a cohort of 362 patients. We evaluate various methods for histopathology representation learning in their ability to capture salient characteristics of underlying tumour morphologies. Results We demonstrate that while foundation models effectively capture morphological signature, they do not significantly outperform simpler aggregation methods in survival prediction. Our proposed bag of visual words approach achieved a C-index of 0.595 and provided interpretable results, identifying high-risk histological patterns such as adipocyte-rich regions. Notably, multimodal GNN configurations did not yield substantial performance gains over unimodal baselines. Conclusion: This work addresses the role of representation quality in graph-based similarity patient modelling, presenting a framework tailored to multimodal learning with a primary focus on histopathology. More importantly, by benchmarking these representations against end-to-end models, we identify key gaps and assumptions in current literature.

Doctoral researcher: Elisa Castagnari
Supervisory team: Ian Simpson, Ole Eigenbrod, Pinar Wennerberg
Project partner: Roche
Abstract: When patients move between hospitals, clinics, and other care providers, their medical records often stay behind in different computer systems that do not talk to each other. This makes it hard for doctors to access the right information at the right time. A lack of good data sharing, also known as poor interoperability, can delay treatment, increase costs, and sometimes even affect patient safety. The project examines how artificial intelligence (AI) technologies and ontologies can facilitate the seamless exchange of health information between systems. The aim is to read and interpret free-text notes written by clinicians, along with laboratory results and EHR data, and transform them into standardised formats and vocabularies that allow for secure and accurate data sharing. We are mapping out what tools already exist, identifying where the biggest gaps are, and building the foundations for new resources to support this work. In the long run, improving interoperability will help healthcare professionals make faster, safer decisions and give patients more coordinated care.

Doctoral researcher: Tesni Walsh
Supervisory team: Joseph Marsh, Simon Biddie
Project partner: NHS Lothian
Abstract: The human genome can be understood as an instruction manual to build and maintain the human body. Genes are often described as containing instructions for making proteins, which are essential for bodily functions. However, only around 1.5% of the genome contains instructions for making proteins (IHGSC, 2001). The remaining 98.5% of the genome is referred to as the non-coding genome, which does not make proteins. While this was previously assumed to be “junk” or redundant DNA, parts of it are now understood to have important regulatory roles in human biology. Variants are changes in DNA sequence in the genome. Most of the time, these variants have no effect on health. However, sometimes they can cause disease. Identifying disease causing variants is important for understanding and treating disease, through early diagnosis, predicting if medications will work, and designing gene editing tools. Despite the importance of being able to identify which variants will cause disease, it is still particularly challenging to identify if a non-coding variant will cause disease. To help do this, computational tools called variant effect predictors have been developed which predict the effects of variants. Since these tools are used in research and healthcare settings, it is important that they are trustworthy and reliable. Therefore, this project is evaluating different AI-based variant effect prediction methods for non-coding variants. This includes using different tools combined with each other, and comparing results with laboratory experiments which directly measure the effects of variants.

Doctoral researcher: Mario Navarro Veiga
Supervisory team: Andrea Weisse, Thamarai Dorai-Schneiders
Abstract: Antimicrobial Resistance in Klebsiella pneumoniae, a WHO critical-priority bacterium, is driven by increasingly frequent multidrug-resistant strains that can resist last-line antibiotics such as tigecycline and ceftazidime-avibactam. This PhD project investigates the dynamics of resistance across different scales through two complementary studies. The first explores population-level resistance trends using longitudinal MIC data from the ATLAS surveillance database. Tracking of temporal and regional resistance trends revealed a global decline in ceftazidime-avibactam susceptibility, and region-specific increases in resistance to tigecycline and ceftazidime-avibactam in China and India, respectively. India also shows high levels of co-resistance between antibiotics. Ceftazidime-avibactam shows a concerning co-association trend given its role as a broad-spectrum last-line antibiotic. The second study investigates resistance at the molecular scale, using a Random Forest classifier to predict the direct binding sites (RamBoxes) of the global transcription factor RamA. This is done through the tokenization of predicted RamBoxes with extracted positional and structural features, combined with RNA-seq differential-expression data. This results in the learning of a weak biological signal, providing a proof of concept for the refinement of the model to reconstruct the RamA regulon and to build a dynamic model of RamA-mediated resistance.

Doctoral researcher: Anthos Makris
Supervisory team: Heather Yorston, Stuart King, David H. Steel
Project partner: South Tyneside and Sunderland NHS foundation trust
Abstract: Idiopathic full-thickness macular holes (MHs) form secondary to age-related abnormalities of the vitreoretinal interface with a prevalence of up to 3 in 1000 people over the age of 55. They appear as a small dehiscence in the neurosensory retina at the centre of the fovea, a highly specialised part of the human retina responsible for fine acuity and colour vision. Spectral-domain (SD) optical coherence tomography (OCT) imaging allows ophthalmologists to diagnose, classify and measure MHs. OCT is a non-invasive, high-resolution imaging technique that uses infrared light to image the retina in 3D. Macular holes can be effectively treated by closing the hole using vitrectomy surgery. They are one of the commonest indications for vitrectomy surgery accounting for ~4000 surgeries in UK and more than 200,000 globally per annum. Predicting the visual outcome after surgery is important to guide the decision to operate and manage patients’ expectations, as well providing insight into their pathology. Several studies have shown that postoperative visual acuity (VA) is correlated with a variety of measures of macular hole size that can be measured on SD-OCT. Various studies have attempted to precisely predict postoperative VA using manual 2D measurements of MHs and preoperative VA, although their predictive ability has been limited. Three-dimensional automated image reconstruction has improved this ability, but there are no current standards for shape, size, and resolution of OCT imaging data captured by different OCT devices for this task. There are also many qualitative features and subtle alterations in retinal anatomy, for example, associated with chronicity, which may be predictive of acuity outcomes and that are difficult to measure. Additionally, image artefacts related to a patient’s eye movement and media opacity pose a further challenge in developing image informatics methods. Most existing machine learning (ML) and deep learning (DL) approaches have focused on the automated classification of macular diseases, such as age-related macular degeneration (AMD), diabetic macular oedema (DME), and MHs from OCT image data. More recently, some DL approaches have attempted to improve the prediction of VA outcomes using OCT data although these have been very limited and mainly in diseases other than MH. This project will try novel ML/DL approaches guided by clinical knowledge to improve the prediction of VA outcome for patients.

Doctoral researcher: Eleanor Harrison
Supervisory team: Olivia Swann, Sohan Seth, Jonathon Taylor
Project partner: Department for Energy Security and Net Zero
Abstract: Childhood acute respiratory infections (ARIs) can impact lifelong health. Cold, damp homes can increase ARI risk. Home energy efficiency measures (HEEMs), otherwise known as retrofits, increase a home’s energy efficiency making homes easier and cheaper to heat. HEEMs can therefore improve health, reduce health inequalities, and help to achieve net-zero goals by reducing carbon emissions. Many HEEMs work by reducing conventional heat loss thus preventing bi-directional air flow. This can allow damp to be trapped inside, propagation of mould and reduced indoor air quality from retaining respiratory irritants and infectants. As such, HEEMs may inadvertently increase children’s risk of ARIs. This project aims to assess how different HEEMs affect childhood ARIs depending on a home’s contextual factors; local outdoor air pollution, climatic region and indoor solid fuel usage. The project analyses a new retrospective birth cohort of children aged 0-5 in Scotland from 2008-2025 examining linked healthcare and housing data. This project will model the associations between contextual factors and childhood ARIs and utilise causal inference to estimate direct effects. Machine learning will be used to predictively model the ARI outcomes of a retrofit depending on the contextual factors of a home. Finally, contextual factors will be grouped by similarity to provide HEEM guidance for Scottish local authorities.

Doctoral researcher: Mohammad Kouli
Supervisory team: Ewen Harrison, Annemarie Docherty, Catherine Shaw
Project partner: Sibel Health
Abstract: This PhD research project aims to address a critical issue in postoperative care—failure to detect and intervene in complications that lead to preventable morbidity and mortality. These complications, such as bleeding, sepsis, and myocardial injury, often result in delayed intervention due to the current limitations in monitoring technologies and systems. Continuous postoperative monitoring, including with wearable sensors, offers a potential solution by providing real-time, multimodal physiological data (e.g., heart rate, respiratory rate, oxygen saturation, skin temperature, etc.) to alert healthcare teams of deterioration promptly. Despite the promise of wearable technologies, no validated algorithm currently exists to reliably predict patient outcomes based on this data. Previous studies in this field have been limited by small sample sizes and narrow methodologies, and there are no large-scale efforts underway to assess the utility of such multimodal monitoring in the postoperative period. This PhD project will address these gaps by developing predictive AI models that incorporate continuous waveform data, patient history, and other clinical variables to predict adverse events after surgery. Building on the foundations laid by the EMUs (Enhanced Monitoring Using Sensors after Surgery) and ICU Heart studies, this project will use a unique, real-time dataset collected from patients undergoing major surgery in various countries. The aim is to develop advanced machine learning models, such as random forests and neural networks, and explore attention mechanisms and multitask learning. These models will predict multiple complications simultaneously, a highly attractive feature from a clinical perspective. The study will have access to a large, unique dataset, ensuring the robustness of the predictive models. The PhD will also involve collaborations with Sibel Health, an industry partner, and ethical considerations such as bias, privacy, and transparency will be closely monitored. The outcome of this research is expected to have a profound impact, potentially reducing preventable deaths and complications in postoperative care worldwide, with relevance in both high-resource and low-resource settings.

Doctoral researcher: Jo Igoli
Supervisory team: Sjoerd Beentjes, Ava Khamseh, Kenneth Baillie, Leonardo Koeser
Project partner: National Institute for Health and Care Excellence
Abstract: Randomised controlled trials (RCTs) are considered to be at the top of the hierarchy of medical evidence. However, there are challenges with running RCTs. These include ethical considerations (e.g., administering non-gold-standard treatments to patients), practical feasibility (e.g., single- or double-blind studies not being possible for surgical interventions), and differences between an RCT and clinical practice. Furthermore, RCTs are the wrong tools for research questions that seek evidence from daily clinical practice. Hence, there is a need to supplement RCTs or fill in the gaps when RCTs are inappropriate or not feasible with high-quality real-world evidence (RWE) from real-world data (RWD). The Causal Roadmap provides a principled and rigorous framework for answering causal and statistical questions from RWD, producing reliable RWE. It requires careful phrasing of the causal question of interest, being explicit about assumptions – e.g., regarding confounders and missingness – required to answer the causal question from available RWD. The Roadmap offers powerful Target Maximum Likelihood Estimators (TMLE) to perform statistical estimation. TMLE can take advantage of machine learning (ML) and artificial intelligence (AI) algorithms in its Super Learner estimation of nuisance functions. The Causal Roadmap is compatible with Target Trial Emulation (TTE), in which an idealised trial is emulated. In this research, we will apply the Causal Roadmap to neurosurgical RWD where there is lack of RCT evidence. The first causal question of interest, following the Roadmap, is to evaluate which current surgical treatment is most effective for improved health outcome in acute hydrocephalus secondary to aneurysmal subarachnoid haemorrhage (aHCP-aSAH). aHCP-aSAH is a life-threatening condition that affects up to 74% of aSAH patients. It is also among the most treated pathologies by neurosurgeons. aSAH can cause severe morbidity in up to 30% of patients and mortality in up to 47%. Appropriately, this first causal question is in line with the research priorities – of generating RWE and assessing aHCP-aSAH treatment effectiveness – for the The UK’s National Institute for Health and Care Excellence (NICE) regulatory body. The results of this research could trigger an update in NICE’s aHCP-aSAH guidelines.

Doctoral researcher: Hailey Deckers
Supervisory team: Carsten Hansen, Yunjie Yang, Fraser Millar
Project partner: NHS Lothian
Abstract: Cancers, as a result of distinct mutations, lead to the expansion of clonal cell populations that exhibit considerable heterogeneity in epigenetic, physical and transcriptome profiles. This pathogenic capacity enables cancer cell survival under nutrient-poor conditions, infiltration, metastasis and combined poses significant treatment challenges, particularly due to the emergence of therapy-resistant cell subpopulations. Therefore, an ability to analyse and predict cellular responses to therapeutics early, and in rare cell populations would be transformative. Our project aims to identify precisely why, and how cancer cells differentiate from healthy cells by integrating Artificial Intelligence (AI) with cutting-edge bioimaging modalities. We will focus on pleural mesothelioma, a deadly cancer caused by asbestos exposure, characterised by mutations in key tumour suppressors. Using state-of-the-art label-free imaging platforms (Nanolive and Livecyte), alongside confocal high content imaging (Opera Phenix Plus), we will perform high-resolution, multi-parametric analysis to understand the unique cellular dynamics and predict patient responses. This approach will leverage our recently developed isogenic cellular model that closely represents the disease, facilitating the development of personalised treatment strategies. Through continuous refinement and validation of our AI-driven model against a panel of promising drug candidates, this project aims to provide ground-breaking improvements in diagnosing and treating mesothelioma, potentially extending to other cancers with similar genetic disruptions. We anticipate that this ambitious project will enhance opportunities for stratification and early diagnosis, and contribute to the development of future therapeutic strategies.

Doctoral researcher: Nardiena Pratama
Supervisory team: Ajitha Rajan, Paul Brennan
Project partner: NHS Lothian
Abstract: Glioblastoma (GBM) is an aggressive brain cancer with a poor survival rate. It is important that doctors make fast and accurate decisions about treatment to improve the patients’ chance of survival. Currently, it takes weeks for doctors to determine whether the patient will benefit from receiving chemotherapy via laboratory tests. The goal of this doctoral research is to build an artificial intelligence (AI) tool that can help doctors decide which treatment to give the GBM patient in a non-invasive and timely manner. Using brain scans and patient data, the AI tool will learn to determine whether the patient’s tumour will be sensitive to chemotherapy. In the first year, we tested various AI models to explore different configurations of existing models and ways to combine different types of data. Our preliminary results show inconsistent performance across different experiments. We discovered that a specific method of combining data resulted in an improvement in how the model can differentiate between two different tumour states. This foundation aims to provide a baseline as the project progresses towards a complete decision-making tool.

Doctoral researcher: Stephen Binaansim
Supervisory team: Karl Burgess, Diego Oyarzun
Project partner: Dynamic Therapeutics
Abstract: Our body’s internal clock, known as the circadian rhythm, helps regulate many important functions such as the sleep-wake cycle, energy use, and hormone activity. When these natural rhythms are disrupted, they are often linked to severe mental illnesses such as bipolar disorder, schizophrenia, and major depression. Although modern laboratory methods can measure many small molecules in the body over time and study their patterns to understand their impact on disease, existing computational algorithms often struggle to detect these meaningful patterns in real-life health data. This project aims to improve how we study these changing biological patterns in people with bipolar disorder by developing better ways to preprocess and analyse the data. We intend to use advanced computational methods to identify important rhythmic patterns that may be linked to disease. In the long term, this work could help support earlier diagnosis, better monitoring of disease progression, and improved assessment of treatment response, while also deepening our understanding of the biological basis of severe mental illness.

Doctoral researcher: Iva Jankovic
Supervisory team: Filippo Menolascina, Vissarion Fisikopoulos
Abstract: This project addresses the problem of identifying selective multi-target metabolic interventions that reduce tumour viability while preserving healthy cellular function. In genome-scale metabolic models of cancer cells, viability is typically assessed at optimal flux states, although it is determined by the structure of the feasible metabolic region. Existing approaches rely on static optimisation and do not capture how interventions reshape this region or account for their sequential nature. We propose a reinforcement learning framework in which intervention design is formulated as a sequential decision problem over constraint-defined feasible spaces. Structural interventions are evaluated using a geometric measure based on the volume of the feasible region, estimated via sampling. By combining reinforcement learning, game-theoretic modelling, and convex geometry, the project aims to identify selective metabolic vulnerabilities in cancer cell lines as candidate multi-target therapeutic strategies.

Doctoral researcher: Sim Mei Choo
Supervisory team: Sohan Seth, Bruce Guthrie, Ewen Harrison
Abstract: Artificial intelligence is increasingly used in healthcare to predict patient risks and support clinical decisions, but many tools remain in research settings and are not adopted in routine care. They may be too rigid, poorly aligned with clinical reasoning and workflows, or not designed with deployment from the start. This PhD project aims to address these gaps by developing a framework to help electronic health record-based clinical risk prediction become more transparent, reviewable and useful in practice. Instead of acting as a black box that only returns a single risk percentage, the proposed framework aims to produce a structured risk output. For each patient, the model would show the predicted risk, how uncertain that prediction is, what relevant medical knowledge supports or challenges it, and whether the prediction needs further review. The key idea is that the model would not rely only on patterns found in patient records. It would also use trusted medical knowledge to check whether a prediction makes clinical sense. Ultimately, this research seeks to support safer clinical decision-making, post-deployment monitoring and health-system oversight as evidence, practice and uncertainty continue to evolve.

Doctoral researcher: Lachin Soufizadeh Balaneji
Supervisory team: Peter Kind, Paul Rignanese, Arno Onken, Thomas Watson
Project partner: Simons Initiative for the Developing Brain
Abstract: Changes in the SYNGAP1 gene can cause a rare neurodevelopmental disorder associated with learning difficulties, epilepsy, and behavioural problems. However, it is still not fully understood how these genetic changes affect patterns of brain activity. In this project, we analysed electrical brain recordings from rats with and without a change in the SYNGAP1 gene. The animals took part in a fear-conditioning experiment in which they learned to associate a light with a mild foot shock. We used machine-learning methods to investigate two questions: whether brain activity could distinguish animals with the genetic change from unaffected animals, and whether it could distinguish periods when the animals were freezing (staying completely still out of fear) from periods when they were moving. The preliminary results showed that both the animal’s genetic group and freezing behaviour could be predicted accurately from patterns of brain activity. The analyses also suggested that different aspects of brain activity contributed to these predictions in affected and unaffected animals. The next stages of the project will test these features more directly and investigate whether they change when normal Syngap function is restored. This may help identify useful measures of brain function for studying the disorder and evaluating potential treatments.

Doctoral researcher: Nikoo Moradi
Supervisory team: Alfredo Gonzalez-Sulser, Javier Escudero
Project partner: Simons Initiative for the Developing Brain
Abstract: Neurodevelopmental disorders (NDDs) are characterised by cognitive, motor, and sensory deficits that appear in early childhood1 . NDDs including severe autism, intellectual disability, and epilepsy often cooccur in patients, while current therapeutic interventions are very limited. Recent breakthroughs through large-scale transcriptomic studies identified hundreds of mutations linked with NDDs . New medical interventions are in development but quantitative biomarkers able to measure the efficacy of novel treatments for these genetic disorders are desperately needed. A prime candidate imaging technique for clinically relevant biomarkers is EEG, as it can be quickly performed and is relatively inexpensive . However, robust EEG biomarkers utilizing conventional signal processing techniques have not been identified. For this PhD project we propose that artificial intelligence techniques such as supervised machine learning and deep learning may identify effective generalizable EEG biomarkers for NDDs.

Doctoral researcher: Han Jiang
Supervisory team: Ian Duguid, Angus Chadwick, Jessica Passlack
Project partner: Simons Initiative for the Developing Brain
Abstract: We constantly adjust our movements to deal with a changing world. Sometimes the goal we are reaching for moves to a new place, and sometimes the movement we need to reach it has to change. Two parts of the brain are thought to handle these different demands. One, the cerebellum, builds a kind of internal model of how our movements turn into outcomes and uses it to fine-tune them. The other, the basal ganglia, learns by trial and error, trying out different movements to find the new target. Signals from both regions meet at a single point, the motor thalamus, before reaching the parts of the brain that produce movement. What is not yet understood is whether the thalamus blends the two signals together or switches between them depending on what the situation needs. This project uses computer models to study that question. First, we build models that learn in each of these two ways, and a third that combines them, and test them on tasks where the goal moves, the movement changes, or both. This shows what behaviour each strategy produces, and whether combining the two helps or gets in the way. Second, we fit these models to the behaviour of real mice doing the same three tasks, to work out which strategy they actually use. Third, we fit the models to brain activity recordings and compare these with behaviour, to see which regions support which strategy, and how the thalamus combines their signals. Together, this work aims to show how the brain flexibly coordinates different strategies to adapt our movements.

Doctoral researcher: Anna Lindahl
Supervisory team: Chris Wood, David Gally
Project partner: Biophoundry
Abstract: This interdisciplinary project will develop an AI-based pipeline to engineer bacteriophage specificity, moving beyond discovery to active design. Leveraging the "Phrameworks" cell-free assembly platform developed with the external partner, Biophoundry, the student will train ML models to identify highly conserved regions on the bacterial surface. Using AI-based protein design methods, the student will design novel Receptor Binding Domains (RBDs) for the phage tail fibre, which will be assembled using Biophoundry’s proprietary "Trinity" technology for experimental validation. This computational-experimental cycle aims to develop effective antibacterial therapies by generating synthetic phages with a broad host range and a reduced risk of resistance evolution.

Doctoral researcher: Albie Sherred
Supervisory team: Andrew Stanfield, Leena Williams, Peggy Series
Project partner: Simons Initiative for the Developing Brain
Abstract: Sensory perceptual disturbances are a widespread feature of neurodevelopmental disorders. They are linked to stress and maladaptive coping mechanisms, and are a prevalent understudied feature leading monogenically inherited forms of intellectual disability and autism, such as Fragile X Syndrome (FXS) and SYNGAP1. Current strategies for their identification and quantification are primarily based on informant reporting, therefore subjective and vulnerable to misclassification, and there are no available treatments. What are then the key objective biomarkers for sensory sensitivities and what are the neuronal circuit impairments underpinning these features in sensory cortices? Presently, we are conducting a human study testing the use of electroencephalogram and a commercially available Brain Gauge tactile stimulator device to objectively quantify tactile impairments in FXS. We also have EEG data from auditory and visual stimuli in both FXS and SYNGAP1 individuals. This AI4Bi PhD project aims to accelerate transformation of our human data into a novel biomarker of sensory sensitivities by (Aim 1) using machine learning, trained on clinical, behavioural and EEG data to develop a classifier. (Aim 2) Then refine the machine learning classifier to determine which features are most predictive and how acquisition time can potentiallty be shortened for future clinical trials. (Aim 3) In parallel, use the framework of Bayesian inference and computational psychiatry to build mathematical and computer models to identify key diagnostic markers of tactile sensitivities in FXS and identify novel objective bio markers for sensory sensitivities. The novel biomarkers for sensory sensitivities uncovered could then be employed in planned therapeutic research, and the findings may be applicable to similar and associated conditions, increasing impact.

Doctoral researcher: Ann-Kristin Balve
Supervisory team: Ajitha Rajan, Eleonora D'Arnese, Rishi Ramaesh
Project partner: NHS Lothian
Abstract: Deep learning has demonstrated strong performance in medical imaging, yet its clinical adoption remains limited due to the opaque, black-box nature of many models. In high-stakes settings like cancer diagnostics, accuracy alone is insufficient; clinicians need clear, interpretable explanations to ensure patient safety and build confidence in AI-assisted decisions. Therefore, this project, by specifically focusing on breast cancer, will be developing a clinician-driven AI framework for breast cancer diagnosis. It will develop a transparent, explainable, and robust solution to support effective, safe, and trustworthy decision-making in real clinical settings.

Doctoral researcher: Aref Andishgar
Supervisory team: Charlie Lees, Catalina Vallejos
Abstract: SENTINEL will integrate electronic health records, patient-reported outcomes, and wearable signals to predict inflammatory bowel disease outcomes, updating in response to new measurements being observed. The student will introduce multi-omics in collaboration with our industrial partner, Nightingale Health, to further improve predictions. Embedded within Edinburgh’s IBD service and the Lees and Vallejos data-science research groups, the work delivers an interpretable pipeline ready to power SENTINEL’s proactive, EHR-adjacent decision support. The models will be trained on population based local NHS data (Lothian IBD Registry fully integrated with DataLoch; n=10,000 IBD patients) and Danish national registries.

Doctoral researcher: Asemaneh Nafe
Supervisory team: Gulsen Surmeli, Sara Wade, Steven McDonagh
Project partner: Simons Initiative for the Developing Brain
Abstract: At the core of the behavioural abnormalities that manifest in ASDs are deficits in brain-wide connectivity. A lack of coordinated activity has been demonstrated using low resolution functional and structural imaging technologies. The specific manifestations of connectivity deficits at the level of individual neurons and brain areas remain largely unknown, hindering mechanistic understanding of a wide range of neurodevelopmental disorders. A major obstacle to progress is the challenge of investigating neural connectivity and vulnerable neuronal populations in a sufficiently high throughput manner. While the technologies for collecting high-throughput data are now available, application to autism research requires optimization of analytical tools. This project will address this challenge by tailoring AI and machine learning based approaches to develop analysis tools for barcoded anatomy. By applying these to datasets acquired from mouse models of the Fragile X disorder we aim to identify changes in the transcriptomic and projectomic characteristics of neurons.

Doctoral researcher: Christos Petritsis
Supervisory team: Raven Hickson, Peter Kind, Marino Pagan, Angus Chadwick
Project partner: Simons Initiative for the Developing Brain
Abstract: The richness and flexibility of the rat behavioural repertoire make them well suited as models of the cognitive and social aspects of neurodevelopmental disorders (NDDs). Standard laboratory housing drastically reduces opportunities for rats to express natural behaviours, therefore vastly diminishing the behavioural repertoire available to study1. The Habitat was designed to provide an environment that more closely aligns with the ecology of the Norway rat to address this mismatch and provide opportunities to observe the development of adaptive behaviours in their functional environment. Housing in the Habitat results in observable effects on the transcriptome, and living in the Habitat has different behavioural effects in two different models as compared with living in standard housing. Habitat housing also appears to alter behaviour at a micromovement scale, as captured by RatSeq (unpublished data, method described here2). However, little is known about what aspects of the Habitat experience may contribute to these effects on behaviour. The Habitat allows the capture of multiple modes of data (RFID tracking, video, audio, etc.) for characterising the animal’s behavioural repertoire with the ultimate goal of predicting genotype. The overarching goal is to generate testable hypotheses about circuit-level differences between models of NDD and wild-types.

Doctoral researcher: Dónal Heelan
Supervisory team: Thanasis Tsanas, Andrew Horne, Philippa Saunders
Project partner: Roche Diagnostics
Abstract: Endometriosis is a chronic debilitating condition affecting about 10% of women of reproductive age. There is an unmet clinical need to facilitate accurate, timely diagnosis, remote symptom monitoring, and intervention assessments. The project will focus on mining the largest longitudinal multimodal datasets in endometriosis (ongoing data collection from our team as part of two large scale grants) to provide new clinically actionable insights into how self-reports, home-collected biological samples, and data from wearable sensors can facilitate endometriosis telemonitoring.

Doctoral researcher: Hadis Ahmadian
Supervisory team: Rik Sarkar, Chris Wood
Project partner: Oxford Drug Design
Abstract: Off-target action is a major challenge in novel drug design; molecules that bind with the target protein are also likely to bind with other similar proteins and introduce unintended side effects. In this project, we will develop generative models that incorporate the requirement of not binding with off-target proteins. This model will be combined with a pipeline of interpretable analysis to help medicinal chemists understand structural features that reduce off target binding. The project has been co-developed with Oxford Drug Design, who will be the industrial partner and advisor on the project.

Doctoral researcher: Hannah Hashem
Supervisory team: Ajitha Rajan, Siddarth N., Alexander Laird
Project partner: NHS Lothian
Abstract: Immune checkpoint inhibitors (ICIs) have markedly improved survival for several cancers, but safe, effective deployment in the NHS requires better tools and data to optimise use and manage toxicities. A major unmet need is robust biomarkers that distinguish responders from non-responders, predict immune-related adverse events and guide personalised therapy. This project addresses that gap by integrating RNA sequencing, whole-exome sequencing and immunofluorescence imaging to predict treatment outcomes and discover novel biomarkers. It will develop advanced AI-based multimodal learning methods to align and combine these diverse data types, aiming to deliver a more accurate, comprehensive picture of tumour–immune interactions.

Doctoral researcher: Imogen Call
Supervisory team: Ting Shi, Louisa Pollock, Cheryl Gibbons
Project partner: Public Health Scotland
Abstract: This PhD project aims to investigate inequalities in maternity vaccine uptake in Scotland using advanced health data science methods. Leveraging national electronic health records and the DataLoch Respiratory Registry, the study will develop computational models to identify patterns in vaccine delivery and access across demographic and clinical factors. Complementary qualitative research will explore maternal attitudes and healthcare delivery models, particularly among underserved groups. By integrating biomedical, clinical, and behavioural data, the project will create a novel, data-driven framework to inform targeted public health interventions and improve maternal and neonatal health outcomes across diverse populations.

Doctoral researcher: Juan López-Ríos
Supervisory team: Antonia Mey, Andrea Weisse
Project partner: BioAscent
Abstract: This project will develop an agent-based active learning framework that integrates human medicinal chemistry expertise into AI-driven molecular design. By embedding domain knowledge within iterative learning cycles, the project aims to create models that not only predict compound performance but also account for synthesisability and design feasibility. The resulting agent-based “human-in-the-loop” system will enable adaptive compound selection informed by both data and expert reasoning. The outcome will be an interpretable, industrially deployable tool that bridges computational discovery and experimental validation, advancing the translation of AI innovations into real-world drug discovery workflows.

Doctoral researcher: Nicola Barbaro
Supervisory team: Kartic Subr, Chris Wood
Project partner: Adaptyv Bio
Abstract: This PhD project addresses the complexity of designing therapeutic protein binders by developing a novel, interpretable abstract representation of proteins informed by Molecular Dynamics (MD) simulations. Current inverse design methods are challenged by the vast sequence-structure space and computational cost. Our primary aim is to create an abstract framework that significantly streamlines the design process, allowing for fast exploration of the protein sequence space to achieve target properties, including specific binding, stability, and desired immunogenicity. This data-driven abstraction, grounded in molecular dynamics, will capture essential features for accurate and efficient design. The approach will be validated using the therapeutically relevant PD-1 system with AstraZeneca. The aim is to overcome limitations in current design methodologies, paving the way for innovations in targeted drug delivery and biosensing.

Doctoral researcher: Rodrigo Guerra Palma
Supervisory team: Ahmar Shah, Saturnino Luz
Project partner: Public Health Scotland
Abstract: This project will use the Public Health Scotland Unscheduled Care Data Mart (UCD)—a linked patient-level dataset covering all of Scotland since 2011—to improve the efficiency and equity of unscheduled care. We will map patient pathways across NHS 24, ambulance, emergency, acute, and mental health services using descriptive statistics, pathway visualisation, and machine learning. Predictive models will identify factors affecting outcomes, while clustering will reveal common pathways and bottlenecks. Embedded within PHS, the project will deliver actionable policy recommendations to enhance data collection, optimise patient flows, and guide equitable redesign of urgent and unscheduled care services.

Doctoral researcher: Tassella Isaac
Supervisory team: Alex Kwong, Ahmar Shah, Heather Whalley
Abstract: Depression is a complex, multifaceted disorder with profound developmental, social, and biological impacts. This project applies data-driven and AI methods to longitudinal youth data to uncover key factors shaping short- and long-term depression pathways. It tackles three challenges: (1) the heterogeneity of depression, including distinct subtypes such as persistent and adolescent-onset forms; (2) identifying the most important predictors among vast biological, psychological, and social data; and (3) capturing short-term changes using ecological momentary assessment and wearable technology. Through AI and machine learning, the project aims to characterise heterogeneity, identify predictive factors, and inform personalised, real-time interventions for youth depression.

Doctoral researcher: Vighnesh Bantwal Kamath
Supervisory team: Nathalie Rochefort, Arno Onken
Project partner: Simons Initiative for the Developing Brain
Abstract: Children on the autism spectrum (ASD) differ from typically developing children in many aspects of their processing of sensory stimuli. One proposed mechanism for these differences is an imbalance in higher-order feedback to primary sensory regions, leading to an increased focus on local object features rather than global context. The aim of this project is to use mouse models to reveal the neuronal encoding of contextual information in the visual cortex. We will determine how visual feedback processing may be disrupted in mouse models of ASD. This project will leverage artificial intelligence tools to uncover how natural stimuli statistics are encoded and how local neuronal populations form both reliable and context-dependent representations of natural scenes in the primary visual cortex. In order to investigate these mechanisms, we will use a combination of large-scale high throughput recordings of neuronal activity and artificial neuronal networks. By using electrophysiological recordings with high density silicon probes, we will record neuronal responses to natural scenes in all layers of the primary visual cortex (V1), in awake head-fixed adult mice. We will use a recently developed modelling framework to generate optimized surround images and movies in order to systematically investigate the rules that determine contextual excitation versus inhibition in a naturalistic setting. The approach is based on a new type of deep learning data-driven model that can accurately predict V1 responses to new (unseen) stimuli. The experimental design integrates large-scale neuronal recordings, a model capable of accurately predicting responses to diverse natural stimuli, the in silico optimization of non-parametric images and movies, and in vivo validation of the predictions.

Doctoral researcher: Levin John
Supervisory team: Nikhil Hirani, Kevin Dhaliwal, Eleonora D’Arnese, Sohan Seth
Abstract: Fibrotic interstitial lung diseases (fILDs) are a heterogeneous group of disorders characterised by progressive extracellular matrix deposition, irreversible lung remodelling, respiratory failure, and premature mortality. Current therapies, including immunosuppressive and antifibrotic drugs, slow disease progression but do not halt or reverse fibrosis and are frequently associated with significant side effects. Prognosis remains highly variable and difficult to predict, particularly in clinically important subgroups such as patients with co-existing chronic obstructive pulmonary disease (COPD), lung cancer, or acute exacerbations of disease, all of which are associated with increased mortality and are under-represented in existing predictive frameworks and therapeutic studies. This project aims to develop a multimodal AI/ML framework integrating longitudinal clinical, imaging, and multi-omic datasets to address prognostic uncertainty and identify therapeutic targets in fILD. Aim 1 focuses on predictive modelling in approximately 2,500 deeply phenotyped patients with MDT-confirmed fILD, incorporating serial pulmonary function tests, treatment history, hospitalisation events, laboratory data, and quantitative HRCT imaging features derived from automated radiomic pipelines. Genomic data from approximately 1,000 subjects will support polygenic risk modelling and genotype-phenotype analyses. The project will specifically model disease progression and adverse outcomes in fILD associated with COPD, lung cancer, and acute exacerbations. Aim 2 will use a deeply phenotyped subset of approximately 200 patients with GWAS, bronchoalveolar lavage (BAL) transcriptomics, BAL proteomics, and serum proteomics to identify molecular pathways and candidate therapeutic targets associated with fibrosis progression. The framework will integrate multimodal data using advanced machine learning approaches within a secure Trusted Research Environment with embedded computational infrastructure. Overall, the project aims to establish a genetics-informed multimodal framework for precision medicine in fILD, improving risk stratification and supporting biologically informed therapeutic discovery.