Supervisors

Professor Robert Stewart

Department of Psychological Medicine, Institute of Psychiatry, Psychology and Neuroscience

Email: robert.stewart@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/professor-robert-stewart

Dr Christoph Mueller

Department of Psychological Medicine, Institute of Psychiatry, Psychology and Neuroscience

Email: christoph.mueller@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/christoph.mueller.html

Background: Many people with mental disorders are prescribed multiple medications for other conditions, many of which have unwanted and potentially harmful side-effects. Due to the complexity of treatment regimens and multiple conditions being managed, it is difficult for clinicians to have an overview of all the potential impacts of polypharmacy. The Medichec application is a decision support tool which highlights potentially harmful side effects, including central anticholinergic action, drowsiness, dizziness, bleeding risk, electrolyte disturbances, constipation, and ECG abnormalities. Medichec has recently been expanded in its functionality and has already been successfully embedded in routine clinical services.

Novelty and Importance:  This project will be the first evaluating outcomes of common medication side effects in large clinical populations of mental health service users. The student will work towards integrating information on medication side effects into the range of dashboards currently under development to support clinical care. This will generate novel interventions ready for early feasibility evaluations. Although likely to focus on older patient populations (including those with dementia) where polypharmacy most commonly occurs (and is associated with higher risks of side-effects), we anticipate taking a disorder-agnostic approach to maximise potential applicability of findings in mental healthcare, aware that polypharmacy and its problems may have an impact at any age.

Primary aims:  Using the Clinical Record Interactive Search (CRIS) system, a state-of-the-art electronic health record (EHR) data analytic platform, the student will:

Examine the impact of known side effect profiles on patient outcomes including mortality, hospitalisation, and cognitive impairment/decline and variations in impact by mental health condition.

Carry out nested matched studies in samples of patients prescribed specific medication groups, to determine whether receipt of medication with known side-effects is associated with more adverse outcomes compared to receipt of other medications from the same class (e.g., antidepressants with bleeding risk compared to antidepressants without bleeding risk).

Integrate this information on medication side-effects into informative dashboards for clinical feedback and evaluate their application.

Planned research methods and training provided:  The student will gain skills in epidemiology and advanced statistical analysis, including techniques to simulate randomized trials and applied research using health records data. The CRIS team and Centre of Translational Informatics infrastructure have supported more than 200 research publications in mental health clinical informatics.

Objectives / project plan

Year 1: Literature reviews and analyses for Objective 1.

Year 2: Analyses for Objective 2 and input to dashboard design

Year 3: Dashboard functionality implementation and early-stage evaluation; thesis preparation.

Two representative publications from supervisors:

Publication 1:  Mbazira, A, Bishara, D, Perera, G, Rawlins, E, Webb, S, Archer, M, Balasundaram, B, Shetty, H, Tsamakis, K, Taylor, D, Sauer, J, Stewart, R & Mueller, C. Sedation-associated medications at dementia diagnosis, their receptor activity, and associations with adverse outcomes in a large clinical cohort. Journal of the American Medical Directors Association 2022; 23: 1052-1058. https://doi.org/10.1016/j.jamda.2021.12.038

Publication 2:  Bishara, D, Perera, G, Harwood, D, Taylor, D, Sauer, J, Stewart, R & Mueller, C. 'The Anticholinergic Effect on Cognition (AEC) Scale - Associations with mortality, hospitalisation and cognitive decline following dementia diagnosis. International Journal of Geriatric Psychiatry 2020; 35: 1069-1077. https://doi.org/10.1002/gps.5330

Supervisors

Professor Richard Dobson

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: Richard.j.dobson@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/richard-dobson  Twitter: @richdobson

Dr Zina Ibrahim

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: zina.ibrahim@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/zina.ibrahim.html

Arijit Patra, Phil Scordis

UCB Pharma UK

Project Details

Background:  Electronic Health Records (EHRs) hold detailed longitudinal information about each patient's health status and disease progression, the majority of which (~80%) are stored within unstructured text. This data provides the opportunity to transform care through learning from the data available on other patients.

Rare neurological conditions are challenging in that they often have diverse clinical presentations, with a relatively small and variable number of patients in different healthcare systems. This project will develop Machine Learning (ML) workflows for automated diagnosis, prognosis and modeling of trajectories associated with rare neurological disorders, leveraging EHRs and natural language processing, and accounting for limited and imbalanced data environments.

The work may also leverage the emerging field of Digital Twins, the idea that we can use AI and large amounts of digital data within EHRs to accurately mimic real-world patients allowing users to model possible scenarios and outcomes on the twin’s real-world counterpart. As the patient data landscape is increasingly multimodal, models that can aggregate multiple data streams in EHR data are desirable.

Characterisation of the rare neurological conditions will leverage existing tools already developed and deployed within partner Trusts such as CogStack, a platform which has a near real-time feed from the EHR system within KCH (>1m patients), SLaM and GSTT, MedCAT for Named Entity Recognition and Linking of the text, and Foresight, a novel transformer-based pipeline that uses a GPTv2/3 language modelling approach to structure and organize EHRs and anticipate a range of future medical events such as:

Predict the risk of diseases; 2) Simulate patient future; 3) Suggest diagnoses; 4) Suggest medications or procedures for multiple conditions.

Novelty and Importance:  The project will develop Machine Learning (ML) workflows for automated diagnosis, prognosis and modeling of trajectories associated with rare neurological disorders, leveraging Electronic Health Records (EHRs).

Primary aim(s):

Planned research methods and training provided:  Software development, Natural Language Processing, Data Analysis, Patient engagement, Research communication.

Objectives/project plan

Year 1: Exploratory data analysis and scoping; literature review; define problem statements and disorder priorities.

Year 2: Build ML workflows for analytics on EHR for problem statements defined in Year 1; Explore initial results; insights to clinical trial design.

Year 3: Write thesis and communicate key findings to stakeholders.

Two representative publications from supervisors:

Publication 1:  Kraljevic, Z, Searle, T, Shek, A, Roguski, L, … & Dobson, RJB 2021, 'Multi-domain clinical natural language processing with MedCAT: The Medical Concept Annotation Toolkit', Artificial Intelligence in Medicine, vol. 117, 102083.

Publication 2:  Bean, D., Kraljevic, Z., Shek, A., Teo, J.T. and Dobson, R., 2022. Hospital-wide Natural Language Processing summarising the health data of 1 million patients. medRxiv.

Keywords:  Health informatics; Natural Language Processing; Artificial Intelligence; Electronic Health Records; Machine Learning.

Maudsley BRC research themes

Child Mental Health and Neurodevelopmental Disorders

Informatics

Trials, Genomics and Prediction

Neuroimaging

Experimental Medicine and Novel Therapeutics

Supervisors

Dr Latha Velayudhan

Department of Old Age Psychiatry, Institute of Psychiatry, Psychology and Neuroscience

Email: Latha.velayudhan@kcl.ac.uk

Website: Latha Velayudhan - Research Portal, King's College, London (kcl.ac.uk)

Dr Christoph Mueller

Department of Psychological Medicine, Institute of Psychiatry, Psychology and Neuroscience

Email: christoph.mueller@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/christoph.mueller.html

Project Details

Background:  People with dementia are known to have increased use of physical and mental health care due to both medical comorbidities and neuropsychiatric symptoms. Besides affecting the individual with dementia, this has a substantial impact on caregivers, social care, and health services. People with dementia are frequently admitted to hospitals and often discharged into residential care settings.  While a proportion of risk factors for developing dementia is modifiable and thereby a target for intervention, little is known about what influences physical and mental health care resource use after receiving a diagnosis of dementia. Knowing these risk factors could feed into the development of support tools similar to the Lester adaption of the cardiometabolic health resource available for people with schizophrenia. Using a large primary care and mental health care data source, we aim to investigate the trajectories and the predictors for physical and mental healthcare use after dementia diagnosis and develop a high-risk for health care use checklist.   

Novelty and Importance:  Modifiable risk factors associated with increased physical and mental health care use in people with dementia will be identified, and the potential for early intervention explored.

Primary aims

To investigate the trajectories of medical and psychiatric problems following a dementia diagnosis, health care resources accessed, and predictors of these presentations.

To devise a high-risk for health care use checklist, embed this in the dementia dashboard currently in development, and test its feasibility and acceptability.

Planned research methods and training provided:  Data will be acquired from the Clinical Record Interactive Search (CRIS) system, which is linked to primary care (via Lambeth Data Net) and hospital-use data (via hospital episode statistics). More than 200 papers in the field of clinical informatics in mental health have been supported by the CRIS team, which will help the student to become proficient in using big data resources.

Objectives / project plan

Years 1 and 2: Data acquisition from CRIS and linked data sources. Analyses on health service use in people with dementia and its predictors.

Years 2 and 3: Development of the high-risk of health care us checklist, embedding it in the dementia dashboard, and evaluation of its feasibility and acceptability.

Year 3: Finishing data analyses; dissemination; write-up and submission of PhD thesis.

Year 4: Submit/revise publications arising from the project; apply for further research funding building on the PhD; allow time for unplanned extensions

Two representative publications from supervisors:

Publication 1:  Mueller, C, Perera, G, Broadbent, M, Stewart, R & Velayudhan, L 2022, 'A retrospective analysis of patient flow in mental health services for older adults in South London during the COVID-19 pandemic', International Psychogeriatrics, vol. 34, no. 3, pp. 297-298. https://doi.org/10.1017/S1041610221002775

Publication 2:  Couch, E, Mueller, C, Perera, G, Lawrence, V & Prina, M 2021, 'The association between an early diagnosis of dementia and secondary health service use', Age and Ageing, vol. 50, no. 4, pp. 1277-12

Keywords:  Dementia; Health care trajectories; Multimorbidity; Electronic health records; Applied clinical informatics.

Maudsley BRC research themes

Informatics

Supervisors

Dr Nicholas Cummins

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email:  nick.cummins@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/nick.cummins.html

Dr Sara Simblett

Department of Psychology, Institute of Psychiatry, Psychology and Neuroscience

Email: sara.simblett@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/sara.simblett.html

Professor Dame Til Wykes

Department of Psychology, Institute of Psychiatry, Psychology and Neuroscience

Project Details

Background:  Each year, 1.4 million people attend emergency departments in England and Wales with a recent head injury. Of these, one-fifth have features suggesting skull fracture or evidence of Traumatic brain injury (TBI). Depression is commonly experienced by up to a third of individuals but is sometimes difficult to accurately detect due to reliance on self-report measures and clinical interviews that are confounded by difficulties with retrospective recall due to memory impairments and difficulties with self-awareness and may delay an accurate diagnosis.

Novelty and Importance:  Novel research from our group in the field of digital phenotyping has been working towards validating the early detection of depression through acoustic and prosodic voice markers. People with depression tend to have reduced pitch variability and intonation and increased phonation and articulation errors. For people whose self-report and recall are problematic, for example following a TBI, this information may be particularly useful. However, to date, no studies have investigated the acceptability and feasibility of such an approach.

Primary aims

To assess the feasibility and acceptability of remote data collection of auditory speech samples and symptoms of depression (‘mood data’) with people who have experienced a TBI.

To investigate whether feasibility and acceptability is different for people from different sociodemographic backgrounds and with different levels of disability, for example, with and without speech impairments after TBI.

Planned research methods and training provided:  Training in the design of ecological momentary assessment (EMA) of speech and AI in healthcare. There will also be a strong emphasis on opportunities to learn about patient and public involvement (PPI) in research, which will involve qualitative methods and an opportunity for a secondment to learn about user experience and hands on speech and mood data collection from a digital technology company. 

Objectives / project plan

Year 1: Systematic review on a topic relevant to the research aims, alongside PPI work (e.g., focus groups) to understand the needs of this clinical group when gathering speech and mood data, with an FTE 0.2 secondment in industry over six months.

Year 2: Data collection for a feasibility and acceptability study gathering speech and mood data using EMA from a diverse sample of people with TBI.

Year 3: Analysis of feasibility and acceptability of gathered speech samples and mood data, stratifying by sociodemographic factors and severity of disability, including speech impairments.

Year 4: To complete write up of a PhD thesis.

Other notable aspects of the project:  The aim would be to work towards the development of a clinical trial following this PhD to analyse how depressed mood relates to speech signals in everyday life and evaluate how speech interventions alter depression trajectories.

Two representative publications from supervisors:

Publication 1:  Cummins N, Dineley J, Conde P, et al. Multilingual markers of depression in remotely collected speech samples. Research Square; 2022. DOI: 10.21203/rs.3.rs-2183980/v1.

Publication 2:  Simblett, S., Matcham, F., Siddi, S., Bulgari, V., di San Pietro, C. B., López, J. H., Ferrão, J., Polhemus, A., Haro, J. M., & de Girolamo, G. (2019). Barriers to and facilitators of engagement with mHealth technology for remote measurement and management of depression: qualitative analysis. JMIR mHealth and uHealth, 7(1), e11325.

Keywords:  Speech; Depression; mHealth; Traumatic brain injury; Neuropsychology.

Maudsley BRC research themes

Psychosis and Mood Disorders

Informatics

Supervisors

Professor Daniel Stahl

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: Daniel.r.stahl@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/daniel.r.stahl.html

Professor Paolo Fusar-Poli

Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience

Email: paolo.fusar-poli@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/en/persons/paolo-fusarpoli(ced1ec59-cb83-4efe-b9f6-5c41302558a7).html

Dr Dominic Oliver

University of Oxford, Warneford Hospital

Project Details

Background:  Approximately 1% of the UK population suffers from psychotic disorders, such as schizophrenia. The onset of psychotic illness in young people can profoundly impact the life course of a young individual and current treatments offer minimal help. It is important to develop clinical prediction models (CPMs) to predict the risk of developing psychoses in individuals with mental health problems to enhance early intervention efforts. This project will focus on the development and validation of individualized CPMs and risk stratification in patients at high risk for psychosis using machine-learning methods

Novelty and Importance:  Current CPMs to predict the risk of developing psychoses are static using only potential predictors collected at the first visit to the mental health hospital. This project aims to develop a dynamic prediction model that automatically updates with the availability of new information. This decision tool would allow monitoring persons at risk and to offer them early intervention.

Primary aim(s):  This project aims to transform our static first episode psychoses risk CPM into a dynamic prediction model that automatically updates risk when new patient information (e.g., treatments, side effects, symptoms) becomes available using modern machine learning methods.

Planned research methods and training provided:  The candidate will establish a pipeline to extract and process electronic patient health records data and then develop a prediction model using machine learning methods to predict the risk of developing psychoses in a dynamic way. Two approaches will be compared: i) a statistical modelling approach (regularized cox landmark model), which is easily interpreted and performs automatic variable selection and ii) dynamic survival random forest models, which allow for the implementation of more complex models at the expense of both interpretability and built-in variable selection. The final model will be implemented in a web-based clinical application. Acceptance among clinicians and service users will be assessed in a feasibility study.

The student will gain relevant training in statistical methods, machine learning, and health informatics from the Department of Biostatistics and Health Informatics established education program.

Objectives/project plan

Year 1: Literature review about dynamic prediction model and early onset of psychoses, preprocessing of data, the establishment of service user group to guide planning of modelling approach and implementation.

Year 2: Development of prediction models using machine learning methods.

Year 3: External validation of models using other mental health national datasets.

Year 4: Implementation into web-based application, small implementation and acceptability study, thesis write up.

Two representative publications from supervisors:

Publication 1:  Irving J, Patel R, Oliver D, Colling C, Pritchard M, Broadbent M, Baldwin H, Stahl D, Stewart R, Fusar-Poli P. Using Natural Language Processing on Electronic Health Records to Enhance Detection and Prediction of Psychosis Risk. Schizophr Bull. 2021 Mar 16;47(2):405-414.

Publication 2:  Fusar-Poli, P., Rutigliano, G., Stahl, D., Davies, C., Bonoldi, I., Reilly, T., & McGuire, P. (2017). Development and validation of a clinically based risk calculator for the transdiagnostic prediction of psychosis. JAMA Psychiatry, 74(5), 493-500. https://doi.org/10.1001/jamapsychiatry.2017.0284

Keywords:  Precision medicine; Dynamic prediction model; Machine learning; Psychosis; Risk prediction.

Maudsley BRC research themes

Psychosis and Mood Disorders

Trials, Genomics and Prediction

Supervisors

Dr Charlotte Tye

Department of Psychology, Institute of Psychiatry, Psychology and Neuroscience 

Email: charlotte.tye@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/charlotte.tye.html

Dr Michael Absoud

Department of Women's and Children's Health, Faculty of Life Sciences and Medicine

Email: Michael.absoud@kcl.ac.uk

Website: https://www.kcl.ac.uk/lsm/research/divisions/wh/groups/reproductive%20biology

Project Details

Background:  Epilepsy is commonly associated with conditions affecting children’s development, which can impact on learning, social and everyday life skills, leading to poor life outcomes and long-term mental health problems with high health service utilization. While research has indicated potential risk factors for early-onset epilepsy and its neurodevelopmental outcomes, most studies have been from primary care databases with limited data quality and uncontrolled retrospective investigations, after diagnosis of neurodevelopmental conditions has been established.

Novelty and Importance:  eLIXIR combines maternal and child electronic health records across south London boroughs, providing a unique opportunity to explore the maternal factors associated with early epilepsy and to determine infant health and developmental outcomes. Linkage to objective neurophysiological (EEG) reports and parent-reported behaviour has not been performed. It is critical that prospective studies are performed to assess early-life risk factors and predictors prior to the emergence of developmental difficulties to target intervention and improve longer-term outlook.

Primary aim(s): 

To investigate the association between maternal and perinatal factors and early-onset epilepsy

health outcomes for mothers of children with early-onset epilepsy

child health and neurodevelopmental outcomes for early-onset epilepsy and link with identified risk factors.

Planned research methods and training provided:  Using eLIXIR infrastructure (approval granted), the student will combine maternity and neonatal data (BadgerNet and eREDBOOK/health visitor records), primary healthcare data (Lambeth DataNet), mental health (CRIS) and link with case ascertainment via EEG from the Evelina London Children’s Hospital, as well as questionnaire and video data securely shared via the vCreate platform (see Section 7). This data will be further linked with participants enrolled in the BEE Study who undergo a range of prospective assessments at multiple timepoints in the first two years of life, capturing information on developmental ability, neurocognition and emerging neurodevelopmental difficulties. This deep-phenotyping will be combined with linkage of maternity and neonatal health records to test findings from the eLIXIR cohort.

The student will be trained in collecting infant behavioural, clinical and neurocognitive measures in BEE, in addition to large-scale data linkage and integration of digital technologies into healthcare records with longitudinal data analysis, to provide a unique interdisciplinary skillset.

Project plan:

Year 1: Training, literature review, vCreate Neuro training/secondment, Aim 1 analysis.

Year 2: Data collection, Aim 2 analysis.

Year 3: Complete infant data collection, Aim 3 analysis.

Year 4: Dissemination of findings, thesis completion.

Two representative publications from supervisors:

Publication 1:  Tye, C., Runicles, A., Whitehouse, A., & Alvares, G. (2019). Characterising the interplay between autism spectrum disorder and comorbid medical conditions: an integrative review. Frontiers in Psychiatry.

Publication 2:  Tye, C., McEwen, F., Liang, H., Underwood, L., Woodhouse, E., Barked, E.D., Sheerin, F., Higgins, N., Yates, J.R.W., TS 2000 Study Group, Bolton, P. (2020). Long-term cognitive outcomes in tuberous sclerosis complex. Developmental Medicine and Child Neurology, 62(3), 322-329.

 Keywords:  Epilepsy; Infants; Development; Autism; Data linkage.

Maudsley BRC research themes:

Child Mental Health and Neurodevelopmental Disorders

Informatics

Neuroimaging

Supervisors

Dr Chloe Wong

Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience 

Email: Chloe.wong@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/chloe-wong

Professor Gerome Breen

Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience 

Email: Gerome.breen@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/gerome-breen

Project Details

Background:  Eating disorders (ED) affect ~8% of the global population (Galmiche et al., 2019). ED are often chronic and cause substantial costs. ED are complex with both genetic and environmental causes. Recent efforts have identified eight genome-wide significant loci to date (Watson et al., 2019) and multiple environmental factors (Larsen et al., 2021). Epigenetics, biological mechanisms that underlie the interaction between genes and the environment, might play significant roles in the aetiology and manifestation of ED, are currently understudied (Hübel et al., 2019). To address this research gap, we propose to study the epigenetic basis of ED using nanopore DNA methylation sequencing data in 4,000 participants from the Eating Disorders Genetics Initiative United Kingdom UK dataset (EDGI UK; edgiuk.org), recently funded by NIHR (a £4 million grant). The project will be supervised by Dr Chloe Wong, an expert in epigenetics and methods, and Prof Breen, and international psychiatric genetics expert and chief investigator of EDGI UK.

Novelty and Importance:  This will be, by 50 times, the largest epigenetic study (i.e. differential DNA methylation) across multiple types of ED disorder diagnoses (see Figure 1 for an UPSET plot of the description of the lifetime diagnoses from EDGI UK participants).

Primary Aim(s):  

The overarching aim of this project is to identify differential epigenetic, i.e. DNA methylation, signatures associated with different types of Eating Disorders and related phenotypes using next-generation Nanopore epigenetic sequencing data from 4000 individuals.

Objectives / project plan:  The team has extensive links with ED charities and Lived Experience; you will also work with them, coproducing the research wherever possible, as part of ongoing participant and public engagement for EDGI UK research.

Objective 1) Nanopore DNA methylation sequencing data of 4000 EDGI will be generated by the BRC BioResource team at the SGPD, IoPPN. The student will be involved in performing fundamental data processing and QC, and pipeline establishment.

Objective 2) Conduct epigenome-wide association study (EWAS) to detect epigenetic signatures of eating disorders.

Objective 3) Conduct epigenome-wide association study (EWAS) to detect epigenetic signatures of extreme BMI phenotypes and lifetime minimum and maximum BMI, as well as other extreme behaviours such as purging.

Planned research methods and training provided: Epigenome-wide DNA methylation data will be generated by the BRC BioResource lab technicians and data processing plus QC will be performed using established pipeline in R. Relevant data analyses training will be provided by the first and second supervisors’ teams.

Year 1: Perform background reading and write a literature review on “Epigenetics in Eating Disorders”. Undertake relevant data analytic trainings and perform data QC and processing on Nanopore DNA methylation sequencing data. 

Year 2: Conduct epigenome-wide association study (EWAS) to detect epigenetic signatures of eating disorder diagnoses (Objective 2).

Year 3: Analyse extreme BMI phenotypes and weights (Objective 3) and to perform sensitivity analyses using single diagnostic groups of no other eating disorders/BMI-related comorbidities.

Year 4: Complete analyses and writing up of papers and thesis.

Two representative publications from supervisors:

Publication 1:  Alameda L., Trotta G., Quigley H., Rodriguez V., Gadlrab R., Dwir D., Dempster D., Wong C.C.Y.*, Forti M.D.* (2022) Can epigenetics shine a light on the biological pathways underlying major mental disorders? Psychological Medicine. *Joint senior authorship.

Publication 2:  Preprint: Dina Monssen, Helena L Davies, Shannon Bristow, Saakshi Kakar, Susannah C B Curzons, Molly R Davies, Zain Ahmad, John R Bradley, Steven Bright, Jonathan R I Coleman, Kiran Glen, Matthew Hotopf, Emily J Kelly, Abigail R Ter Kuile, Chelsea Mika Malouf, Gursharan Kalsi, Nathalie Kingston, Monika McAtarsney-Kovacs, Jessica Mundy, Alicia J Peel, Alish B Palmos, Henry C Rogers, Megan Skelton, Brett N Adey, Sang Hyuck Lee, Hope Virgo, Tom Quinn, Tom Price, Johan Zvrskovec, Thalia C Eley, Janet Treasure, Christopher Hübel, Gerome Breen. The Eating Disorders Genetics Initiative (EDGI) United Kingdom medRxiv 2022.11.11.22282083; doi: https://doi.org/10.1101/2022.11.11.22282083 

Keywords:  Epigenetics; Eating Disorders; Nanopore Sequencing; BMI; Whole genome sequencing.

Maudsley BRC research themes:

Eating Disorders and Obesity

Trials, Genomics and Prediction

Supervisors

Professor Gerome Breen

Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience 

Email: Gerome.breen@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/gerome-breen

Dr Jonathan Coleman

Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience 

Email: jonathan.coleman@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/jonathan.coleman.html

Project Details

Background: There is an urgent need for new drugs in psychiatry, with new modes of action and fewer side effects. Genome-wide association studies (GWAS) in psychiatry have been enormously successful. They have the potential to restart largely paused psychiatric drug development pipelines. Industry does not currently have methods to directly translate and analyse (poly)genetic data to identify potential new therapeutic drugs. Remarkably, very expensive and highly error prone laborious manual assessment by biologists, chemists, and geneticists is still required/preferred for each GWAS locus.

The new Psychiatric Genomics Consortium MDD GWAS (unpublished) has identified >500 genetic variants associated with depression in nearly 500,000 depression cases, significantly enriched for the targets of approved antidepressants. Notably, while the additional power provided by broadly defined depression cases is valuable, antidepressant target enrichment was primarily found in cases meeting full MDD criteria.

Novelty and importance: This studentship will use the largest genetic datasets available for broad depression and narrowly defined MDD to directly identify drug repositioning opportunities and small molecules, leading, with functional and therapeutic validation to new clinical trials in MDD.

Primary aim(s) and Methods:

Aim 1: To conducted sophisticated GWAS of a homogenously phenotyped mega-cohort of individuals with major depressive disorder (estimated 100K affected, >300K unaffected individuals), comprising several cohorts with highly consistent phenotyping.

Aim 2: Using data mining and machine learning frameworks, the student will extend our existing drug targetor pipeline, focusing on MDD and related psychiatric disorders.

Aim 3: To make this into a systems biology and neuroscience-informed AI for drug repurposing of approved drugs and discovery of small molecules.

Planned research methods and training provided:

In aim 1, this studentship will uniquely be able to make use the Genetic Link to Anxiety and Depression (GLAD study) and new data from UK Biobank, Dutch and Australian collaborators to (a) work to conduct a genome-wide association study (GWAS) of major depressive disorder (MDD) adjusted for specific confounders; (b) use these results to re-weight MDD GWAS meta-analyses from international consortia and (c) use these reweighted meta-analyses to identify drugs and small molecules for MDD. The next aims will develop one layer for hypothesis generation (the AI) and one layer for experimental validation, which will be subsequently iteratively used by the AI. In this project we will work with neurobiologist Prof Robert Hindges (MRC Centre for Neurodevelopmental Disorders), who is separately funded to carry out screening of MDD associated genes in zebrafish models. The first supervisor, Prof Breen, is the PI of the GLAD study and an internationally-recognised expert in psychiatric genetics, with a strong interest in leveraging GWAS for drug discovery. The second supervisor, Dr Coleman, is a statistical geneticist with extensive experience in the conduct of GWAS, and leads an emerging research group using GWAS to develop empirically-testable hypotheses in neurobiology.

Anticipated timeline:

Year 1: Training for Aim 1 and begin to undertake GWAS; training in machine learning and AI frameworks for Aims 2 and 3.

Year 2: Complete and publish Aim 1 GWAS. Reweight GWAS from international consortium, apply to existing pipeline as part of Aim 2.

Year 3: Complete and publish Aim 2. Begin to develop pipeline for Aim 3, incorporating published empirical data on systems biology and the brain epigenome.

Year 4: Complete Aim 3, integrating empirical data from collaboration with Prof Hindges. Write thesis.

Two representative publications from supervisors:

Publication 1:  Coleman JRI, Gaspar HA, Bryois J; Bipolar Disorder Working Group of the Psychiatric Genomics Consortium; Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium, Breen G. The Genetics of the Mood Disorder Spectrum: Genome-wide Association Analyses of More Than 185,000 Cases and 439,000 Controls. Biol Psychiatry. 2020 Jul 15;88(2):169-184. doi: 10.1016/j.biopsych.2019.10.015. Epub 2019 Nov 1. PMID: 31926635; PMCID: PMC8136147.

Publication 2:  Gaspar HA, Gerring Z, Hübel C; Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium, Middeldorp CM, Derks EM, Breen G. Using genetic drug-target networks to develop new drug hypotheses for major depressive disorder. Transl Psychiatry. 2019 Mar 15;9(1):117. doi: 10.1038/s41398-019-0451-4. PMID: 30877270; PMCID: PMC6420656.

Keywords:  MDD; Depression; Drug discovery; Statistical genetics; Bioinformatics.

Maudsley BRC research themes:

Eating Disorders and Obesity

Trials, Genomics and Prediction

Supervisors

Professor James Teo

Department of Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience

Email: jamesteo@nhs.net

Website: https://kclpure.kcl.ac.uk/portal/james.teo.html

Professor Mark Edwards

Department of Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience

Email: mark.j.edwards@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/mark-edwards

Project Details

Background:  Over 50% of people with disease or damage affecting the nervous system have psychiatric symptoms. These symptoms are rated by patients as their most difficult problems, and they predict higher healthcare costs, morbidity and mortality. In addition, functional neurological disorder, a neuropsychiatric condition, is one of the commonest diagnoses made in neurology outpatients.  Despite this clear need, access to treatment and research activity in neuropsychiatric disorders are limited. In the latest Neurological Alliance survey of over 6000 patients with neurological illness, nearly 70% indicated that their mental health needs were not met.

Novelty and Importance:  Progress in developing improved pathophysiological understanding, treatments and services for people with neuropsychiatric disorders is hampered by key knowledge gaps: prevalence estimation, impact, determinants of outcome and service use, and patterns of comorbidity.  We need to identify patients, their common comorbidities, and to identify areas of high need within a particular neurological diagnosis and for specific psychiatric diagnoses in a transdiagnostic fashion.

With the unique possibilities offered by the informatics and AI engine behind Cogstack, and the integration of Cogstack within Kings College Hospital, the Maudsley Hospital Guys and St Thomas’ Hospitals and General Practitioner records, we have the opportunity for the first time to track and correlate psychiatric symptoms in those with neurological illness and Functional Neurological Disorder.

Primary aim(s):  

To determine the prevalence of psychiatric disorders, their associated health outcomes and influence on patterns of healthcare utilization and costs in people with neurological disease and those with functional neurological disorder.

Planned research methods and training provided:  Training in clinical informatics, data querying in Python, population health measurement, health economics. A unique opportunity to use the latest machine learning and clinical informatics tools in the NHS.

Objectives / project plan

Year 1: Determine the prevalence of neuropsychiatric disorders using NHS informatics systems at Kings College Hospital and Guys & St Thomas Hospitals.

Year 2: Perform comparative studies of healthcare utilization, healthcare costs, medication and receipt of illness-related financial benefits in populations with specific neurological diagnoses with and without psychiatric symptoms.

Year 3: Informatics-based cohorting of FND patients through co-occurrence of related clinical codes of functional somatic symptoms suggestive of undiagnosed or early FND; experimental approaches include graph-based detection relative to previous psychiatric and neurodevelopmental diagnoses.

Year 4: Work with patient groups and other stakeholders to generate research priorities for service and treatment development in people with neuropsychiatric disorders.

Two representative publications from supervisors:

Publication 1:  Hospital-wide Natural Language Processing summarising the health data of 1 million patients; Daniel Bean, Zeljko Kraljevic, Anthony Shek, James Teo,  Richard Dobson; medrxiv preprint. https://doi.org/10.1101/2022.09.15.22279981

Publication 2:  O'Keeffe, S., Chowdhury, I., Sinanaj, A., Ewang, I., Blain, C., Teodoro, T., Edwards MJ,  Yogarajah, M. (2021). A Service Evaluation of the Experiences of Patients With Functional Neurological Disorders Within the NHS.. Front Neurol, 12, 656466. doi:10.3389/fneur.2021.656466

Keywords:  Neuropsychiatry; Functional Neurological Disorder; Informatics; Phenotyping; Healthcare Utilization.

Maudsley BRC research themes

Informatics

Supervisors

Dr Ewan Carr

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience 

Email: ewan.carr@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/ewan-carr

Professor Kimberley Goldsmith

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: kimberley.goldsmith@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/kimberley-goldsmith

Project Details

Background:  30% of people in the UK have multiple long-term conditions, accounting for over half of hospital admissions. Long-term conditions are characterised by fluctuating symptoms with periods of remission followed by relapse. Conventional techniques cannot reflect the complex, longitudinal dynamics of symptom trajectories featuring non-linear progression through successive disease stages or treatments.

This project will investigate alternative techniques that leverage rich clinical data to build a detailed picture of symptom fluctuations during the management of long-term conditions. This includes (1) Generalised additive models, a flexible approach to uncovering hidden longitudinal patterns; (2) Gaussian process modelling, a probabilistic model for complex processes; and (3) multistate models describing progression through disease states.

Novelty and Importance:  The digital transformation of healthcare is generating ever larger amounts of data that are high dimensional, high frequency and multimodal. This includes routine assessments in electronic patient records and remotely-collected information from smartphones and wearable devices.

New data streams offer tremendous potential to understand and thereby improve patient outcomes, but only if paired with appropriate statistical methodology. This PhD will bring new insights into prognostic trajectories to inform treatment adaptations and referrals.

Primary aim(s):  

To evaluate, apply, and develop methods to uncover the dynamics of patient outcomes during treatment and management of long-term conditions.

Planned research methods and training provided:  The student will receive advanced training, including:

Joint models for longitudinal data(Netherlands Institute for Health Sciences)

Prediction modelling(King’s College London)

Statistical methods for prognostic models (University of Birmingham)

Objectives / project plan:  

This project exploits existing infrastructure and data, including:

Integrating Mental & Physical healthcare: Research, Training & Services(IMPARTS) has, since 2012, developed infrastructure to monitor patient-reported mental/physical health at Guy’s and St Thomas’ Hospital (GSTT) and King’s College Hospital (KCH).

Remote assessment of disease and relapse in major depressive disorder (RADAR-MDD), a longitudinal cohort study (n=623) with information collected via smartphones and wearable devices over 24 months (e.g., physical activity, sleep).

Year 1 Objectives:

Systematic review; simulation study comparing chosen methods.

Also:

Data access and cleaning; launch event; pre-registration; training.

Year 2 Objectives:

Application in real-world data; publish simulation study.

Also:

Training; present to patient groups.

Year 3 Objectives:

Publish applied studies; write thesis.

Also:

Patient dissemination event.

Year 4:  Write thesis; seek postdoctoral funding.

Any other notable aspects of the project: This project builds upon existing BRC infrastructure and provides a platform for future translation with patient-centred digital tools.

Two representative publications from supervisors:

Publication 1:  Skelton, Carr et al. (2022) “Trajectories of depression and anxiety symptoms during psychological therapy for common mental health problems” Psychological Medicine (forthcoming). https://psyarxiv.com/8scpx/

Publication 2:  Matcham, Carr, et al., (2022) “Predictors of engagement with remote sensing technologies for symptom measurement in Major Depressive Disorder”. Journal of Affective Disorders 310 (1). doi: 10.1016/j.jad.2022.05.005

Keywords:  Logitudinal; Trajectories; High-dimensional; Routine data; Dynamic.

Maudsley BRC research themes:

Psychosis and Mood Disorders

Informatics

Trials, Genomics and Prediction

Supervisors

Dr Angus Roberts

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: angus.roberts@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/angus-roberts

Dr Tao Wang

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience

Email: tao.wang@kcl.ac.uk

Website: https://kclpure.kcl.ac.uk/portal/tao.wang.html

Professor Fiona Gaughran

Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience

Project Details

Background:  Identifying eligible participants for a clinical trial requires matching a person’s detailed medical information with the complex criteria of the trial. Although electronic health records (EHR) have facilitated access to patient data, most information for accessing trial eligibility (e.g. family history) is recorded as free text and cannot easily be retrieved through a database query. Clinicians/researchers often need to manually examine clinical text to identify potential participants, which is time-consuming and expensive.

Natural language processing (NLP) techniques have been used to identify a trial cohort from EHR text. However, existing systems often rely on document-based information extraction, without summarized patient-level information across multiple documents. Thus, complex trial criteria (e.g., “young females who are not trying to become pregnant”) cannot be queried seamlessly, which limits their usability. Also, without integrating multimodal data (both structured and unstructured information) to form a holistic and unified representation space for patients, these systems cannot provide insights into the quality of a cohort and the impacts of various trial criteria.

Novelty and Importance:  Leveraging ongoing research on document-based information extraction in the BRC and established C4C infrastructure at SLaM, this project will develop a clinical knowledge graph (CKG) that links medical entities (e.g., patient, diagnosis and treatment) extracted from both structured and unstructured EHR data with existing medical ontologies (e.g., SNOMED), to enable reasoning over complex trial criteria, and assessment of the representativeness of a cohort. Moreover, candidates who had shown early symptoms of a disease but had not been diagnosed can be identified, given their high similarities to diagnosed peers, which would improve trial safety when the disease is a key exclusion criterion. This project will help address the long-standing recruitment challenge in mental health clinical trials and provide an important basis for further development in this understudied area.

Primary aim(s): 

Develop a multimodal, automated cohort identification method and evaluate against existing methods.

Clinical use cases will align with BRC priorities, including treatment-resistant schizophrenia and clozapine; and mental health trials identified from the Centre for Innovative Therapeutics.

Planned research methods and training provided:

Natural language processing

Network science

Deep learning

Training:  Training on NLP, Clinical Informatics, Deep Learning, Network Modeling and Statistical Methods through KCL courses.

Objectives / project plan

Year 1: Review literature; define data schema for patient-trial knowledge graphs.

Year 2: Develop prototype for a trial cohort identification and retrieval system.

Year 3: Finalize prototype development; evaluations with stakeholders.

Year 4: Dissemination, model sharing.

Two representative publications from supervisors:

Publication 1:  Wang, T., Bendayan, R., Msosa, Y., Pritchard, M., Roberts, A., Stewart, R. and Dobson, R., 2022. Patient-centric characterization of multimorbidity trajectories in patients with severe mental illnesses: A temporal bipartite network modeling approach. Journal of biomedical informatics, 127, p.104010.

Publication 2:  Kraljevic, Z., Searle, T., Shek, A., Roguski, L., Noor, K., Bean, D., Mascio, A., Zhu, L., Folarin, A.A., Roberts, A. and Bendayan, R., 2021. Multi-domain clinical natural language processing with MedCAT: the medical concept annotation toolkit. Artificial Intelligence in Medicine, 117, p.102083.

Keywords:  Clinical informatics; Data science; Natural Language Processing; Clinical knowledge graph; Trial cohort identification.

Maudsley BRC research themes:

Psychosis and Mood Disorders

Informatics

Trials, Genomics and Prediction

Experimental Medicine and Novel Therapeutics

Supervisors

Professor Richard Dobson

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience 

Email: richard.j.dobson@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/richard-dobson

Dr Nicholas Cummins

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience 

Email: nick.cummins@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/nicholas-cummins

Project Details

Background:  Speech is uniquely placed in health marker; no other signal contains its singular combination of cognitive, neuromuscular and physiological information. Models developed in speech studies have the real potential to provide unique preventive and predictive information about health to provide opportunities for enhanced self-management or screening services. These advantages aside, the potential of speech as a digital phenotype is yet to be realised.

Novelty and Importance:  A core reason behind lack of translation of speech phenotypes is a lack of replication analysis and robust generalised testing of the predictive power of potential speech phenotypes across health conditions. Existing speech toolboxes (E.g., openSMILE1 and Praat2) are not suitable for enabling this need on a wide scale. Neither were designed for clinical applications and do not contain extraction methodologies relating to a basic standard speech parameter set specific for capturing health information. Both require considerable computer science knowledge to reliably operate. Therefore, to realise the potential of speech phenotypes there is an urgent need for an open-source speech features extraction toolbox specifically designed for clinical applications.

www.audeering.com/research/opensmile/

https://www.fon.hum.uva.nl/praat/

Primary aim(s): 

This thesis will have two main aims:

Develop a basic standard speech parameter set, informed by Patient and Public Involvement (PPI), specifically for capturing health information.

Package all developed code as open-source (Python/R) libraries, which researchers and clinicians can easily import to enable the analysis of similar data.

Planned research methods and training provided:  The student would undertake the Core Research Skills training programme offered through KCL. They would also be able to undertake relevant modules as part of the Applied Statistical Modelling and Health Informatics programme offered by the department of Biostatistics and Health Informatics. There will also be a strong emphasis on opportunities to learn about PPI in research.

Objectives / project plan

Year 1: Conduct a systematic review of speech parameters typically used in health analysis alongside PPI work to understand what feedback clinical groups want from speech signals.

Year 2: Develop code to extract identified key speech parameters. Conduct a 6-month 0.4 FTE secondment with industry to further develop coding skills.

Year 3: Utilise existing speech database to conduct statistical analysis over a range of typical speech-health problems to highlight the strength of the developed features.

Year 4: Package code into open-source format and write up of PhD.

Two representative publications from supervisors:

Publication 1:  Ranjan Y, Rashid Z, Stewart C, Conde P, Begale M, Verbeeck D, Boettcher S, Dobson R, Folarin A, RADAR-CNS Consortium. RADAR-base: open source mobile health platform for collecting, monitoring, and analyzing data using sensors, wearables, and mobile devices. JMIR mHealth and uHealth. 2019 Aug 1;7(8):e11734.

Publication 2:  Cummins N, Dineley J, Conde P, et al. Multilingual markers of depression in remotely collected speech samples. Research Square; 2022. DOI: 10.21203/rs.3.rs-2183980/v1.

Keywords:  Speech; Digital Phenotyping; mHealth; Signal Processing; Feature Extraction Techniques.

Maudsley BRC research themes:

Child Mental Health and Neurodevelopmental Disorders

Psychosis and Mood Disorders

Informatics

Supervisors

Dr Alfredo Iacoangeli

Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience 

Email: Alfredo.iacoangeli@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/alfredo-iacoangeli

Professor Ammar Al-Chalabi

Department of Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience

Email: ammar.al-chalabi@kcl.ac.uk

Website: https://www.kcl.ac.uk/people/ammar-al-chalabi

Project Details

Background:  Patient’s heterogeneity, both in terms of clinical presentation and of biological causes of the disease, are factors that greatly affect the effectiveness and design of clinical trials. For example, longer trials are necessary to evaluate the impact of a therapy on disease progression for people with a slow-progressing form of the disease with respect to people with a fast-progressing form, and people with the same clinically defined disease, but different underlying mechanisms, might respond differently to the same treatment. This scenario is well exemplified in people affected by neurological and neuropsychiatric disorders such as Schizophrenia (SCZ) and Motor Neurone Disease (MND/ALS) which are characterised by highly variable clinical manifestations and a plethora of pathogenic mechanisms. In our laboratory, we have analysed biological (omics) and clinical data of thousands of SCZ and MND patients using machine learning and statistical approaches and identified subtypes of patients with distinct clinical outcomes and underlying candidate mechanisms. This PhD will apply such models to classify patients into homogenous classes and evaluate the differences in the progression of the disease in longitudinal datasets and their potential for the stratification of patients in clinical trials.

Novelty and Importance:  Patient heterogeneity is a factor that affects the effectiveness of all clinical trials and therapy development. Multi-omics machine learning (ML) approaches have shown great potential to provide a higher degree of resolution into the complexity of MND and SCZ. However, interpretability and reproducibility have limited their use. We will apply, for the first time, robust, validated and replicated methods that we have developed in the recent years, to large longitudinal datasets and for the re-analysis of trial data.

Primary aim(s):  

To evaluate the usability of ML approaches based on multi-omics data for the prediction of diseases progression in longitudinal studies and for the stratification of patients in clinical trials.

Planned research methods and training provided:  Training in Machine Learning, Bioinformatics, omics techniques and cluster computing will be provided by the supervisors’ groups and via the attendance of specific courses from the postgrad programme in Applied Statistical modelling and Health Informatics of the Department of Biostatistics and Health Informatics.

Objectives / project plan:

Year 1: In the first year the student will undergo training in machine learning, high performing computing and bioinformatics. The student will then gain access to the multi-omics datasets currently stored on KCL facilities and use the analysis framework and methodology developed in the laboratory for the subclassification of patients in these datasets. Currently available datasets include RNA-sequencing, DNA sequencing, Methylation, Proteomics and clinical data of MND, SCZ and controls cohorts (approximately 800 individuals) from Project Mine, KCL BrainBank and CommonMind Consortium.  

Year 2: In the second year the student will get access to longitudinal multi-omics datasets and use the analysis framework from year 1 to identify patients subgroups in these cohorts. They will then investigate whether the modelled subgroups present distinct progression patterns. Longitudinal data of MND and SCZ patients with matching omics, are available for a subset of the Project MinE dataset and from the PsyCourse study via collaboration.

Year 3: The third year will be dedicated the re-analysis of clinical trial data using the outputs from year 1 and 2. The student will test whether the different patient subtypes responded differently to treatment and whether this information can add value to the evaluation of the trial results. Multi-omics and trial data from are currently available on KCL facilities from the Mirocals study.

Year 4: This final year will be dedicated to the optimization of the classification model using feedbacks from the results obtained in year 2 and 3, the release of the method as open-access software on Github, and to the thesis and article writing.

Two representative publications from supervisors:

Publication 1:  Tam, Oliver H., et al. "Postmortem cortex samples identify distinct molecular subtypes of ALS: retrotransposon activation, oxidative stress, and activated glia." Cell reports 29.5 (2019): 1164-1177.

Publication 2:  McLaughlin, Russell L., et al. "Genetic correlation between amyotrophic lateral sclerosis and schizophrenia." Nature communications 8.1 (2017): 1-12.

Keywords:  Neurological disorders; Psychiatric disorders; Clinical trial; Machine learning; Patient stratification.

1. Maudsley BRC research themes:

Psychosis and Mood Disorders

Informatics

Trials, Genomics and Prediction