University of Glasgow: Precision Medicine Projects
University of Glasgow: Precision Medicine Projects
Below you will find an exciting and diverse range of available 18/19 MRC DTP in Precision Medicine projects. The projects listed below are all based at the University of Glasgow.
Information on 'How to Apply' is available here.
A collaborative systems biology-based approach to dissect novel mechanisms of cardiovascular disease
Cardiovascular diseases (CVD) are the major cause of morbidity and mortality worldwide. Whilst some of the key mechanisms of the development of CVD are firmly established the interplay of such mechanisms and the contribution of additional, yet unknown pathophysiological principles is subject to ongoing research. We propose a project that will address this gap in knowledge by employing an interdisciplinary (cardiovascular research / computing science) and translational (preclinical / clinical) approach using systems biology methods. We have generated complex omics datasets in a range of preclinical models and clinical cohorts with cardiovascular diseases that will form the basis for in-depth analysis within this project. Whilst this proposal will train the successful candidate in systems biology and bioinformatics techniques it will be firmly embedded in our active research groups addressing mechanistic and clinical questions in CVD. As such we are reaching out to candidates with a background in statistics, mathematics or computing science who are interested in cross-disciplinary training that will include basic wet lab and clinical research techniques. The project will compare rodent and human datasets and dissect concordant and discordant molecular features that will be subject to further validation and, in the longer term, development of therapeutic approaches.
A machine learning framework to predict virus host-species tropism
Emerging viruses like Ebola, Zika, and Dengue virus are the cause of major global health problems. Most emerging viruses have reservoir host species and/or insect vectors but not all hosts contribute equally to transmission. Identifying which reservoir hosts and arthropod vectors perpetuate transmission cycles is crucial to mitigate disease threats to human health, but the current practice of combining evidence from field surveillance, phylogenetics, laboratory experiments, and whole system perturbations is time consuming and often inconclusive. We have developed supervised machine learning frameworks that predict: 1. the reservoir hosts of twelve key groups of RNA viruses, 2. whether their transmission involves an arthropod vector
and 3. the identity of that vector, directly from viral genome sequences.
In this project, we will use iterative processes of machine learning and in vitro experiments in order to:
i. Extend the computational framework to predict the host associations of dsRNA viruses, which include important human and animal pathogens;
ii. Identify the viral genomics determinants allowing arboviruses to replicate both in mammalian and insect cells and digitally design host-optimized genomes;
iii. Use reverse genetics in in vitro assays in order to produce host-optimized genomes and assess the predictive power of the developed computer framework.
This project will address fundamental questions in virus-host interaction using artificial intelligence and modern virology techniques.
Applying "big data" techniques to clinical trials in heart failure to inform prognosis
In randomised trials, the repeated clinical, symptom-related and quality of life measurements made at study visits at different time points are often not fully modelled which is both an inefficient use of the data collected, and could lead to important prognostic information being overlooked. Over the last decade, there has been growing interest in the joint modelling of longitudinal and survival data that are collected during randomised trials as they have a number of important advantages over traditional methods. In this PhD project, the aim is to identify distinct recovery/improvement trajectories among heart failure trial patients, and determine what baseline patient characteristics are associated with this. Such information would be valuable for clinical decision-making. Further, this research will explore whether treatment influences the trajectories, and, in turn, the association with survival. Such findings identified for population subgroups would have particular importance for trials that produced overall null effects.
As well as involving exciting and important research, this PhD will train the student in key data science skills using software and version control favoured by experts. Further, the student will develop expertise in statistical modelling and in developing web-based apps to improve the translation of results to clinical audiences.
Are gamma delta T cells associated with prognosis in colorectal cancer
Colorectal cancer (CRC) is the second most common cause of cancer death in Europe. Although outcomes have improved over the past decades, survival still remains poor, with 5-year survival of 50% across all stages of disease. Immune cell infiltration, particularly CD8 T-cells, is an independent prognositic indicator of good outcome for CRC patients. However, very little is known about other immune cell types, such as gamma/delta T cells, in this regard. Gamma/delta T-cells are gut-resident cells involved in tissue homeostasis that make-up a large proportion of T-cells in the gastrointestinal tract. Studies from mouse models indicate that gamma/delta T-cells can be either pro- or anti-tumorigenic. Therefore, the overall goal of this project is to examine the relationship between gamma/delta T-cell infiltration and patient survival, cancer subtype as well as commonly dysregulated signalling pathways in CRC. The project will involve analysing human patient samples by immunohistochemistry and various gene expression methods using robust statistical analyses. At the end of the study, the student will have established a link between gamma/delta T-cell abundance in CRC and patient outcome. These data may identify new pathways to target therapeutically and uncover new patient subgroups that are sensitive to gamma/delta T-cell immunotherapy.
Assessing the value of high-priced novel treatments in oncology in an era of precision medicine: methodological and policy tools
Health systems around the world are struggling to accommodate the increasing cost of new cancer treatments. In the UK a new version of the cancer drugs fund has recently been initiated and both the European Society of Medical Oncology (ESMO) and the American Society of Clinical Oncology (ASCO) have recently published ‘value frameworks’. This PhD will explore both methodological challenges in showing value for new oncology products (such as: use of surrogate endpoints; the need for extrapolation; biomarker/subgroup analysis; adjusting for cross-over) and policy tools designed to improve the value of these new treatments to the health systems (such as: cancer drugs fund in UK; patient access schemes; outcomes-based contracting in US; use of value frameworks, application of stratified/precision medicine). A unique feature of this PhD will be the opportunity to collaborate not only within the University of Glasgow, but also with researchers at Memorial Sloan Kettering Cancer Center in New York through international collaborative links with that organisation. Research into the value of cancer therapies is timely and this PhD represents a real opportunity to address a contemporary health care policy problem from an international perspective.
Cryo-EM computational data processing to reveal mechanisms of essential mitochondrial complexes of Toxoplasma gondii as means to develop anti-malarials
Dr Lilach Sheiner
Dr Alexey Amunts
Mitochondria are the energy and nutrient providers of almost all eukaryotic cells. Since different organisms live in varying environments with different nutritional needs, mitochondria are divergent between organisms. Much of the work on mitochondrial biology has been focused on a small group of model organisms, thus current data base are poorly annotated and based mostly on distant homology models that results in little insight into mitochondrial diversity. In the case of mammalians vs apicomplexan (the parasites that cause toxoplasmosis and malaria) the differences can be leveraged for anti-parasitic drug development. For example, a protein complex of the mitochondrial electron transport chain (mETC, essential for making energy) that is different between apicomplexans and humans is the basis for the activity of the anti-malarial Atovaquone. Despite this promise, the structure-function relationship of the apicomplexan mETC complexes are unknown. We will combine the most recent advances in cryo-EM to enable atomic structure determination of the targeted complexes and use this information to re-annotate the current database. This will allow not only accurate dissection of the structure-function relationship of complexes in the mETC, potentially providing knowledge for future rational drug design, but also produce a new reliable source of information for the entire research community.
Data mining and characterisation of dark metagenomic sequence data
The advent of high-throughput sequencing (HTS) has led to an increasing deluge of metagenomic sequence data being deposited in online archives. However, a significant proportion of sequence data cannot be classified as it displays no homology to any known sequence, and such sequence data is typically discounted from further analyses. The aim of this PhD project is to develop a complete computational framework for the data mining of online archives for dark sequence data, combined with the assembly, storage, clustering, and initial classification of such dark sequences. The project aims to answer fundamental questions related to the extent and diversity of dark sequences, classify these sequences into related groups, and predict their biological function and origin. There are 3 broad stages: (1) The development of data mining pipelines to automatically retrieve meta and sequence data from the short read archive. (2) The adaptation of existing metagenomic assembly pipelines towards sequences of unknown origin, and the development of a database system to store and query the assembled dark sequences. (3) The quantification, analysis and clustering of the identified dark sequences. The project is a combination of data science and bioinformatics, with substantial elements of computation, programming and statistics/machine learning.
Disentangling viral respiratory infections and coinfections: a modelling approach that links population surveillance and experimental data
Multiple viruses are responsible for respiratory infections and co-infections that cause a range of disease, from a mild cold to life-threatening pneumonia. Respiratory viruses are generally studied as single entities and not as a community. However, evidence suggest that there are interactions between viruses that are important for patient health and for infection dynamics at the population scale. Mathematical models are essential tools for the control of infectious diseases and have been used extensively at the population scale. However, only a handful of studies have focused on the individual host scale and even less attention has been paid to coinfections. This project will use data from experimental infections and coinfections to model virus-virus interactions. We will combine quantitative data on virus entry, virus replication, virus spread, innate immune kinetics and cellular death and regeneration. Results from this work will identify the relative roles of direct (virus-virus) and indirect (virus-immune response) interactions in coinfections with the goal of improving patient care. We are seeking a candidate with a quantitative background and an enthusiasm for modelling biological processes to join a group of laboratory and quantitative scientists within CVR and The Boyd Orr Centre for Population and Ecosystem Health.
Establishing a therapy-responsive FOXO1 gene signature in chronic lymphocytic leukaemia
Chronic lymphocytic leukaemia (CLL) is the most common blood cancer in the UK, with ~3,500 new UK diagnoses/year, and remains incurable. Although the majority of patients initially respond to current treatments, all eventually relapse due to re-emergence of leukaemic cells that escaped treatment. There is a critical need to establish biomarkers that enable discrimination between treatment-responsive and non-responsive CLL patients. CLL cells interact with the lymphoid organ microenvironment, which provides survival and growth prompts to the leukaemic cells, resulting in disease progression. Healthy cells have checkpoints that prevent over-expansion of cells, however these checkpoints are disrupted in cancer cells. Forkhead box, class O (FOXO) proteins can behave as molecular brakes in cells, regulating processes responsible for cell cycle arrest and promoting apoptosis. We show that FOXOs are inactivated in CLL cells through interaction with the tumour microenvironment, resulting in a reduction of the FOXO-mediated transcription programme. Furthermore, drug treatment can re-activate FOXO and induce CLL cell death. We hypothesise that CLL cells isolated from treatment-responsive patients will display active FOXO-mediated biomarkers, which will be lost in patients exhibiting resistance to treatment. Taking an ‘omics’ approach, we aim to elucidate the biomarker signature that accompanies FOXO activation thus providing an important clinical tool to enable clinicians to follow patient responsiveness to treatment, and generate a research tool for identifying novel treatments that target CLL cells, thus assisting in the generation of new treatments for CLL patients.
Integrated modeling of epidemiological variables from imperfect medical records
Health-related data, especially electronic medical record data, are routinely collected as part of clinical practice and there is increasing interest in their potential to address important questions in health services research, for example, whether reporting biases slow time to diagnosis of clinical conditions, such as colorectal cancer or heart failure. The sheer volume of available data is as vast as it is heterogeneous in its completeness and quality. Modern statistical methods developed in other areas of environmental data-science have the potential to interpret, error-correct, or completely reconstruct the heavily distorted epidemiological pictures presented by these platforms-of-opportunity. This project will seek to develop detailed models for complicated observation processes (such as primary care medical records) and tease apart the variables distorting disease incidence, from the underlying risk factors that generate disease, and that may influence primary care referral patterns and ultimately health outcomes.
We are fortunate to have access to large, routinely collected clinical datasets such as the expansive SAIL databank or clinical trial data such as that obtained as part of the Early Cancer detection test – Lung cancer Scotland (ECLS) study (www.eclsstudy.org) that has recruited over 12,000 patients. By examining such anonymized data and the information available in primary health practices this project will look at how information about chronic illnesses and lifestyle factors are recorded in primary care data compared to research data and identify systematic biases, for example, in relation to socio-demographic factors or geography.
Machine- and Deep-Learning Methods for Analysing Complex Health Simulations
Dr Eric Silverman
Dr Claudio Angione
Population health research has made significant strides in recent decades, but some health challenges facing society remain very difficult to study. Issues like the ageing population, increasing obesity, and multi-morbidity are driven not by simple cause-and-effect relationships, but are influenced by behavioural, environmental and social factors. Increasingly, we are turning toward interdisciplinary computational modelling techniques such as agent-based modelling (ABM) to unravel the complex interplay of factors that drive these urgent problems in population health. ABMs are computer simulations that model the behaviours of individual people in complex virtual environments, and consequently help us better understand how the interaction of the individual, the environment, and the social realm lead to poor health.
This project will tackle this exciting area of research head-on by applying ABM to key problems in population health, and using cutting-edge AI techniques to better understand the behavior of these complex models. Machine- and deep-learning methods can improve the theoretical understanding of the ABM, help calibrate the model, and facilitate interpretation of the results relevant to end users. The PhD candidate will develop novel frameworks for the analysis of large simulation models using machine learning and deep neural networks. These innovations will facilitate the development of new techniques, protocols and software for the analysis and dissemination of complex simulation studies in population health, opening up new avenues of research on major health challenges.
Mass spectrometry imaging of the gut-brain axis to find novel microbiome metabolites that cross the blood brain barrier
The microbiome is now understood to hold the key to many diseases previously thought not to have a microbial input. Determining the impact of the microbiome in these diseases and harnessing this knowledge may offer new intervention strategies across a variety of diseases.
Using mass spectrometry imaging (MSI) we have for the first time mapped the gut-brain axis in a mouse model, determining the presence and distribution of both neurotransmitters and novel, potentially microbiome-derived, molecules alike. MSI is a unique technique that works independently of tagging and allows the building of 2D and 3D maps of tissue sections. Using this data, and through training at state of the art MSI facilities at AstraZeneca in Cambridge, this PhD seeks to exploit the molecular information contained within this dataset to identify novel microbiome derived compounds that cross the blood brain barrier. These compounds will then have their structures solved and where possible through further analysis, have a function ascribed to them. As a proof of principle one such microbiome derived compound is already identified and its functionality currently being tested. This interdisciplinary PhD will provide training in MSI, microbiology and cell biology techniques.
Mathematical Modelling of Haematopoiesis and transmission of malaria
Malaria is caused by protozoan parasites (Plasmodium spp) that require a mosquito vector for transmission. We have recently demonstrated that many of the specialised forms of the parasite required for transmission (gametocytes) originate from the sites of red blood cell production (bone marrow and in the mouse, the spleen). In order to learn more about this process we propose to enhance our current quantification of gametocyte production through photoswitchable-based lineage tracing and mathematical modelling of the subsequent data. This will be coupled to transmission rates to identify the source and timing of production of optimally transmissible gametocyte. Such gametocytes will then be characterised in greater detail using state of the art nucleic acid sequencing technologies. We expect to learn when the best gametocyte is made and what it looks like. This knowledge will form the basis for the informed generation of measures to prevent transmission by the fittest gametocytes. The project represents an opportunity to work in a world-class Centre for Parasitology research (Wellcome Centre for Molecular Parasitology) developing state of the art technologies addressing a problem of global importance.
New artificial intelligence methods for exploring the interaction between diet, metabolome and microbiome in Crohn's disease
At the heart of the current understanding of the cause of Crohn's disease is the interplay between diet and gut microbes. Current metabolomics (for diet) and genomics (for gut microbial) measurement technology have the potential to reveal this interplay but the computational techniques to help discover these relationships do not exist. In this PhD, the student will work with experts in CD, food industry, nutrition, clinical medicine, metabolomics analysis, and metagenomics analysis to build the computational models, based upon artificial intelligence and machine learning techniques, that will help to identify these relationships and further our understanding of CD and the efficacy of current treatments.
Pan-Cancer analysis of histological images with machine learning powered automatic image process pipelines
Histological image is one the primary diagnostic tools in cancer. Encoding morphological and behavioral features of large numbers of cancer and normal cells, analysing histological images could comprehensively characterise the tumour microenvironment in large patient cohorts. Current manual and computational approaches are inadequate to provide the required deep tissue phenotyping across cancer types, leaving challenges in translating the increasingly well-characterised genetic diversity to phenotypic diversity in tissue.
The aim of this project is to construct a panoramic landscape of image-based cellular morphological features from > 10,000 patients across >30 cancer types. To achieve this goal, we will build image processing pipelines powered by state-of-the-art machine learning algorithms that are capable of automatically analysing large-scale histological images from large patient cohort including The Cancer Genome Atlas and a locally hosted unique colorectal cancer dataset. We will also explore if common morphological features can be extracted from microscopic and radiological images.
The project will be based on a solid foundation of algorithmic and experimental expertise within the team members. The candidate will have the opportunity to develop highly effective machine learning-based image processing pipelines and test them on large scale patient data.
Personalized management of blood testing for renal function in patients with and without diabetes
Testing of renal function using blood and urine samples is a key method of monitoring health of patients with diabetes and is recommended every year. Renal function is also tested ad hoc in people without diabetes. However, the evidence base to support either of these approaches is weak and may well result in unnecessary cost burdens to the NHS. Using Scotland’s world class health informatics, we propose a data science approach encapsulating routinely collected health data in NHS Greater Glasgow and Clyde, with 1.2M registered patients, to develop an evidence based algorithm for personalized renal function testing. We will supplement this with a cohort study of 20,000 adults (Generation Scotland) where we will study the phenotypes of broadly healthy people who get renal function measured in routine practice, using record linkage available in Scotland. Thus the project will provide a holistic overview of who is currently getting renal function tested in practice, and what the optimal approaches for the future would look like. This transformative approach has the potential to influence future clinical guidelines. The student will gain exposures to key quantitative skills and well as developing clinical insights and experience in data synthesis from review of clinical guidelines.
Predictive factor analysis via redundancy clustering with application to mental health outcomes in a clinical high-risk group
While there are many successful applications of machine learning techniques to clinical data sets, these methods tend to focus on prediction performance at the expense of scientific understanding. This project aims to develop a technique based on information theory to obtain predictive factors each consisting of a set of multi-model predictor variables which provide a common influence on the output, and therefore elucidate possible causal mechanisms. This method will be developed within a cutting edge ongoing longitudinal mental health study: Youth Mental Health Risk and Resilience Study (YouR-Study). The YouR-study aims to identify neurobiological mechanisms and predictors of psychosis-risk with neuroimaging, genetics and psychological assessments. Schizophrenia (ScZ) is a severe mental illness with enormous economic and social costs. Clinical high-risk (CHR) criteria have been developed that allow the identification of young people at risk of developing ScZ. While current CHR-criteria are sufficient for diagnosis, they are not sensitive or specific enough to predict psychosis-risk on an individual level, a key objective for a precision-medicine based approach to early intervention. The predictive factors obtained with the new method will help address this fundamental issue, as they will provide optimum multi-modal biomarkers.
Predicting response to therapy in Rheumatoid Arthritis: Computational modelling of targeted transcriptional signatures to enable patient stratification
Rheumatoid Arthritis (RA) is associated with considerable morbidity and premature mortality. The therapeutic options for patients has increased with the arrival of a number of biologic therapies, which are now an integral part of the management of patients with RA. However, there is still a substantial proportion of patient who only partially respond or do not respond to these therapies. The ability to stratify patients a priori into responders and non-responders would be an invaluable clinical tool to define which patients will or will not respond. Using pre-treatment samples acquired in a recent clinical study evaluating response to two biological therapies (ORBIT; Porter et.al., Lancet 2016), we have identified via machine learning approaches two discrete transcriptional/clinical covariate signatures that can stratify patients into responder and non-responders. This raises the possibility that stratification of patients could be achieved using baseline transcriptomic and clinical profiles in order to improve patient outcomes. In order to develop predictive tests for use in routine clinical practice, several steps need to be undertaken. This studentship will focus on (a) translating these translating these transcriptional signatures into a molecular modality (Targeted RNA-seq) that is compatible with clinical laboratory testing and (b) applying machine learning techniques to create new predict models that will eventually enable clinicians to stratify patients prior to therapy.
Proteome-level diversity in RNA viruses
RNA viruses are major threats to human and animal health and account for the majority of emerging infectious diseases. This is in large part due to the highly error-prone replication of RNA virus genomes, which results in rapid evolution. These high mutation rates should also affect viral gene transcription. However, the diversity of protein products that result, and how these could affect infection and immunity, is unclear.
This project will consider influenza viruses, a genus which includes both endemic and emerging threats to human health. Novel data analysis tools will be developed to identify signatures of mutation and variant post-translational modification in viral proteins, using mass spectra collected from purified viruses by the Hutchinson group. Maps of protein-level diversity will be compared to next-generation sequencing of the viral genome and transcriptome, in order to map of sites in viral proteins which can tolerate variation and sites which select against it.
The project will provide the first indication of how the high levels of mutation in an RNA virus genome actually relate to protein polymorphisms. By mapping sites of protein conservation and plasticity it will expand our understanding of targets for the immune response, including by ‘universal’ vaccines, and provide a rational basis for viral protein engineering.
Role of the alternative splicing in the gametocyte production in malaria parasites
Plasmodium is a unicellular parasite responsible for human malaria. During its complicated life cycle, it generates a range of highly adapted life forms based on the relatively small and highly compact genome. For example, inside of the human red blood cell it can develop in male, female or asexual form, all of which are characterised by different functions and morphology. The presence of introns in the high proportion of parasite’s genes and identification of multiple alternative splicing events raises the possibility that the editing of messenger RNAs plays an important role in this process.
The aim of the proposed project is to map the splicing events involved in the parasite differentiation into male and female gametocytes and validate them using the laboratory malaria model, P.berghei. The student will analyse both existing and newly generated RNA-seq datasets from different platforms in order to identify the genes is of interests, and use one of the conditional genome editing system available in P.berghei to study the impact of these events on parasite development.
The application of Artificial Intelligence in Pathology Research to discover, develop, validate and deliver novel diagnostics
The integration of digital technology within Histopathology has facilitated the transfer of valuable image-rich data for the purposes of diagnosis, education and research and will undoubtedly further enable precision medicine. Over recent years, there has been a drive to incorporate quantitative analysis processes into these platforms so that morphological information can be converted into quantitative data, for research and clinical applications. Artificial Intelligence (AI), which involves the utilization of deep learning technologies, has recently been integrated within existing digital pathology platforms. This provides a unique opportunity to utilize image analysis, with AI, as a quantitative digital tool, to assist Pathologists in the complex interpretations of tissue-based sections. This brings the potential to contribute to consistency in diagnosis, assimilate ‘big data’ to permit early detection of minor changes associated with disease processes and, in conjunction with companion diagnostics, enable patient stratification for predicting response to therapies. The aim of the project is for a biomedical scientist or biologist to work closely with the Pathology group and Edwards group in Institute of Cancer Sciences University of Glasgow and OracleBio Ltd, a leading industry-based pathology imaging solutions company, to leverage research using AI to not only determine its value and impact for future pathology investigations and research, but also to assist in the implementation of a novel predictor in gastrointestinal cancer. It is expected that the student will spend up to 50% of their time at OracleBio Ltd. OracleBio will provide image analysis platforms, including AI algorithms, and the support of industry leading quantitative tissue-based expertise. The project will enable development of knowledge of disease processes, histopathology tissue recognition, application of molecular techniques and computer science algorithms used on digital image analysis platforms, within wider research and development and commercial contexts.
The health of looked after children in Scotland
Children who are looked after (at home or away from home) are at risk of poorer educational outcomes and reduced life chances but little is known in Scotland about the relationship between health and being looked after. Although a relatively small group, the number of children looked after has increased substantially in Scotland over the last 20 years. The aim of this study is to examine how the health of school-age looked after children compares to school-age children in the general population of Scotland. A large population data set with individual level data on looked after children, and non-looked after children in Scotland will be linked to a range of health data (including hospital admissions, birth registrations and prescribing data) and other additional important factors (e.g. age, sex, deprivation, parental social class and type of care placement). Data linkage for this project has been approved and is currently underway (expected to be completed in summer 2018). Standardised illness rates in the group of looked after children will be compared to illness rates in the general population of children, across all of Scotland and by deprivation category, and survival models will be used to examine the effects of the additional factors on the duration of time until poor health. Findings will allow greater understanding of the health needs of this important group and will help to deliver better outcomes for looked after children.
Understanding associations between cardiometabolic and brain health using Mendelian randomization in UK Biobank.
Cardiometabolic diseases like obesity, diabetes and hypertension are risk factors for dementia, cognitive impairment and stroke. This comorbidity is common and increasing, with significant clinical and public health implications.
It can be difficult to estimate accurate associations between cardiometabolic risk factors and brain health, because a limitation of cross-sectional data is that measurements are usually taken simultaneously – it is a ‘snapshot’. Links between cardiometabolic risk factors like higher blood pressure, adiposity and cholesterol, and worse brain health (e.g. white matter hyperintensities/integrity, gray matter volumes, cognitive scores) cannot parse out what came first, and to what extent the association is confounded by other variables like smoking.
A method called ‘Mendelian Randomization’ generates more causal estimates of association. It uses genetic mutations known to associate with individual modifiable risk factors (like blood pressure) as proxies for lifetime exposure to that risk factor.
UK Biobank is a population study of whom around 15,000 currently have cognitive, genetic and MRI data (increasing in scanning batches until n=100,000). This project will apply Mendelian randomization to estimate causal associations between cardiometabolic diseases and risk factors, and different metrics of brain health using existing data.
Utilising MRI-derived tissue characteristics and deep learning to segment high-resolution functional brain activity in neuroscientific experiments
Deep learning has transformed many fields of image analysis, and likewise, applying this technology to solve medical image analysis problems holds great promise. However, optimal application of deep learning to MRI analyses of the human brain requires researchers with expertise in multiple areas of study, specifically, MRI physics and neuroscience in order to understand the properties of brain tissues, and machine learning in order to understand what models are appropriate to decipher the complex data produced by different image contrasts in MRI. In the proposed project, the student will develop algorithms for segmenting different layers of human cortex from high-resolution MRI scans. The student will have access to the University of Glasgow’s state-of-the-art MRI facilities and knowledge, computational power, and expertise in human systems neuroscience and computer vision from the labs of Lars Muckli and Alessio Fracasso. With the skills, knowledge and competencies acquired, the student will be well-equipped to approach many types of medical imaging questions in the future, including, for example, applications to tumour and stroke imaging.
Unlocking the power of variation. Using differences between doctors to determine which are the best treatments
Our multidisciplinary team of supervisors has extensive experience in exploiting electronic health records and administrative health data to conduct research; this research has led to many high impact publications, changing clinical practice.
In this project, the successful applicant will develop and validate an approach to determine the effectiveness of treatment using routine healthcare data based around the validation and use of “instrumental variables”. This work will exploit existing expertise and enhanced-routine datasets as well as our in-depth knowledge of the clinical challenges in cardiovascular disease.
The successful applicant will gain experience of collaboration across disciplines, developing expertise in working with large complex datasets. The applicant will also develop skills in academic writing, as well as data cleaning, modelling and visualisation. They will use version control software and literate programming approaches to enable robust and efficient production of reports, presentations and academic papers. These skills will be useful to a candidate seeking to develop an academic career, or to take on a postdoctoral role using quantitative methods in other settings.
Use of biological informatics to provide real time intelligence data on invasive group A streptococcal infections
Interested in joining the fight against emerging global infections? This project will use the evolving pandemic of invasive group A streptococcal (iGAS) infections as a model system to develop Artificial Intelligence systems built with real time data produced by the Scottish Reference laboratory for iGAS. Diseases caused by iGAS include necrotizing fasciitis (referred to in the press as “skin eating” bacteria) and blood stream infections. The increasing volume of data generated by emerging real time technologies such as whole genome sequencing lends itself to the use of more intelligent computer systems and algorithms to analyse past, current and emerging data sets and inform public health action. The volume of data available will allow a wide range of approaches, including state of the art deep learning techniques that can represent multi-modal data (e.g. bacterial phenotypic and genomic data) in a lower dimensional space (preserving complex, non-linear relationships) in which automatic identification of outliers and data visualization is possible. Investigate additional meta-data: e.g., co--infection with respiratory tract viruses and evaluate them to determine if these improve the ability of systems to reduce false positive and increase true positive alert rates. This PhD studentship will allow the applicant to work within a multi-discipline team linked into an established Nationally funded programme of work (Health Protection Scotland iGAS surveillance (Reynolds) and the Reference Laboratory Services (Smith lab), International collaboration with genomic biology of GAS (Musser lab) and Computer Sciences (Hughes lab).
Use of Routinely Collected Health Data to Predict Sudden Death and Other Catastrophic Events
Sudden, often unheralded, death is one of the most common modes of death. Quite apart from the personal tragedy, this often has devastating consequences for families, emotionally and economically. Many diseases of the heart and circulatory system can cause sudden death, either as their first presentation or as a subsequent complication. This analysis will describe the rate of sudden death in people with or without known heart problems in the West of Scotland and develop AI strategies to predict those at increased risk. This information will help inform doctors, policy-makers and future research on strategies to prevent events.
Supervisors will provide medical education about relevant cardiovascular diseases to help put the data and objectives into context. The successful candidate will receive training in data-linkage, analysis, and cleaning and receive hands-on training in modern machine learning techniques (e.g. deep networks, text modelling, etc.). As such, this PhD addresses two of the skills priorities highlighted by the MRC (quantitative and interdisciplinary skills).
Universal methods to analysis clinical data
The Immune-Mediated Inflammatory Disease Bio UK consortium (IMID-Bio UK) is building a database that will hold clinical data as well as “omics” data (transcriptomics, proteomics, metabolomics). This is a paradise for data scientists and a holy grail to understand disease. Not only are rich datasets stored, but it will be possible to compared different studies, with similar omics datasets. This PhD will aim to build models to interpret some of these data and visualize the outcomes. Studies will initially focus on rheumatoid arthritis data generated at the University of Glasgow. In the first instance, different classifiers will be tested on expression data to predict the outcome of treatment (precision medicine). Subsequently, proteomics and clinical data will be integrated into the analysis pipelines. The models will be designed to work on the IMID-Bio database and therefore will be useful to the community to query the data.