Computational models for clinical trial design in Huntington's disease
Lead Research Organisation:
University College London
Department Name: Computer Science
Abstract
Huntington's disease (HD) is a progressive neurological disorder caused by an expansion mutation in the HTT gene, and is characterised by motor, cognitive, and behavioural symptoms. Due to its monogenic and fully penetrant nature, it is possible to identify a 'pre-manifest' group of gene-positive carriers who have not yet developed symptoms, presenting a potentially large window for therapeutic intervention. However, as yet there has been only a single FDA-approved disease-modifying Phase 3 trial, and no late-stage trials that have successfully treated mutant HTT. This is partly due to standard trial outcome measures - such as the unified Huntington's disease rating score and related composite scores - having limited sensitivity for detecting treatment effects. The situation is further complicated by a complex association of motor, cognitive and psychiatric symptoms, which can produce different phenotypes within the same clinical diagnosis group.
Computational disease progression models (DPMs) utilise multi-modal data to provide fine-grained measures of subtype, stage and disease time course. DPMs have a broad range of potential clinical applications:
i) identification of disease-tracking biomarkers from early to late stages;
ii) stratification of diagnosis groups into disease subtypes;
iii) estimation of time between key clinical events, such as brain atrophy and motor onset;
iv) identification of subgroups at target stages.
DPMs have already been applied to a wide range of neurological diseases, including Huntington's disease, various kinds of Alzheimer's disease (sporadic, posterior-cortical atrophy, and dominantly inherited), Alzheimer's disease in Down syndrome, fronto-temporal dementia, and multiple sclerosis. Along with modern machine learning methods, DPMs have the potential to provide personalised disease trajectories using data-driven methods applied to clinical, imaging and genetic markers. Such patient-specific information can be used to provide fine-grained stratification for clinical trials, and potentially aid in treatment planning in a clinical setting.
This Fellowship aims to change the way disease modifying therapies (DMTs) are developed by designing early-phase clinical trials around computational models of disease progression. This Fellowship proposes to address this in a two-fold approach: first, I will redefine phenotypes using
machine learning techniques applied to large multi-centre cohort data. Next, I will use model-based analysis of multi-modal data - properly informed by the underlying biology - to reveal the disease mechanisms driving the observed phenotypes, and identify key biomarkers. With an appropriate knowledge of clinical methodology and trial design, I will then integrate model insights into a new framework of entry and treatment criteria for the next round of Phase 1 and 2 clinical trials. This framework can naturally be extended to Phase 3 trial design, and in more generally to other types of disease.
Computational disease progression models (DPMs) utilise multi-modal data to provide fine-grained measures of subtype, stage and disease time course. DPMs have a broad range of potential clinical applications:
i) identification of disease-tracking biomarkers from early to late stages;
ii) stratification of diagnosis groups into disease subtypes;
iii) estimation of time between key clinical events, such as brain atrophy and motor onset;
iv) identification of subgroups at target stages.
DPMs have already been applied to a wide range of neurological diseases, including Huntington's disease, various kinds of Alzheimer's disease (sporadic, posterior-cortical atrophy, and dominantly inherited), Alzheimer's disease in Down syndrome, fronto-temporal dementia, and multiple sclerosis. Along with modern machine learning methods, DPMs have the potential to provide personalised disease trajectories using data-driven methods applied to clinical, imaging and genetic markers. Such patient-specific information can be used to provide fine-grained stratification for clinical trials, and potentially aid in treatment planning in a clinical setting.
This Fellowship aims to change the way disease modifying therapies (DMTs) are developed by designing early-phase clinical trials around computational models of disease progression. This Fellowship proposes to address this in a two-fold approach: first, I will redefine phenotypes using
machine learning techniques applied to large multi-centre cohort data. Next, I will use model-based analysis of multi-modal data - properly informed by the underlying biology - to reveal the disease mechanisms driving the observed phenotypes, and identify key biomarkers. With an appropriate knowledge of clinical methodology and trial design, I will then integrate model insights into a new framework of entry and treatment criteria for the next round of Phase 1 and 2 clinical trials. This framework can naturally be extended to Phase 3 trial design, and in more generally to other types of disease.
Technical Summary
A major barrier to successful disease modifying therapy (DMT) development in Huntington's disease (HD) is its biological and phenotypic heterogeneity, which makes it difficult to define suitable biomarkers and inclusion criteria for clinical trials. Recent advances in the fields of machine learning and computational modelling have made it possible to infer patient-specific information from large and varied datasets. Such information has the potential to improve clinical trial design, and hence accelerate the development of a DMT for HD.
This project will use a large multi-modal imaging and clinical patient dataset, to:
i) construct a baseline array of computational models for patient stratification, using standard statistical and machine learning methods;
ii) use more advanced machine learning methods to identify potential disease subtypes and model patient-specific trajectories of key macro-scale biomarkers for early-stage clinical trial design;
iii) define subtype-specific models of micro-scale protein dynamics in HD for DMT design.
Specifically, this project will develop methods in subtyping, deep probabilistic modelling, and trajectory modelling that have been successfully used in Alzheimer's disease. This project will use the largest combined multi-modal HD dataset - comprised of three separate cohort studies - to evaluate the potential of machine learning techniques in multi-centre studies, which are essential for DMT verification in a rare disease like HD. The proposed framework will naturally extend to Phase 3 trials, and different types of disease, making it a model for next-generation clinical trial design.
This project will use a large multi-modal imaging and clinical patient dataset, to:
i) construct a baseline array of computational models for patient stratification, using standard statistical and machine learning methods;
ii) use more advanced machine learning methods to identify potential disease subtypes and model patient-specific trajectories of key macro-scale biomarkers for early-stage clinical trial design;
iii) define subtype-specific models of micro-scale protein dynamics in HD for DMT design.
Specifically, this project will develop methods in subtyping, deep probabilistic modelling, and trajectory modelling that have been successfully used in Alzheimer's disease. This project will use the largest combined multi-modal HD dataset - comprised of three separate cohort studies - to evaluate the potential of machine learning techniques in multi-centre studies, which are essential for DMT verification in a rare disease like HD. The proposed framework will naturally extend to Phase 3 trials, and different types of disease, making it a model for next-generation clinical trial design.
Planned Impact
This Fellowship will develop methods in machine learning and computational modelling to gain new insights into Huntington's disease (HD) progression, and to establish a new framework for clinical trial design centred about computational methods. The outputs are improved disease understanding and computational tools for patient stratification and clinical trial design. This will have the following impacts:
i) Development of disease modifying therapies
The primary route to treating HD is through disease modifying therapies, which can slow the progression of the disease. The research conducted in this Fellowship aims to accelerate the development of such therapies, by selecting the most suitable group for testing a given therapy, and hence allowing clinical trials to become smaller (in terms of the number of participants) and shorter (in terms of the length of the trial). The focus of this project is on Phase 2 trial design, but it can naturally be extended to Phase 3, thus facilitating a fully computerised framework that can be used to optimise future trial design. With a current clinical trial for HD on-going, this framework has the potential to be validated using real trial data within the next 3-5 years. Once validated, it can then be applied to other neurological diseases, and ultimately any disease. This will impact on the clinical trials community (academic and industry), and people with HD.
ii) Clinical practice and patient care
The computational stratification tools produced by this Fellowship will help inform clinical decision making, by providing personalised patient information. These tools will help clinicians distinguish between different phenotypes, how these phenotypes are expressed together, and how a given phenotype is expected to progress in time. As part of this Fellowship, I will be trained at the UCL Huntington's Disease Multi-Disciplinary Team Clinic, which will help me ensure that the computational information can be used aid in treatment planning, therapy, and patient care. In the short-term (~3 years), this will impact clinicians in their views on modern computational methods and how they can be used in practice. In the longer term (5-10 years), the methods developed in this study could be used to impact on clinical decision making, and hence the care of people with HD.
iii) Pharmaceutical industry
The computational framework for clinical trial design proposed by this Fellowship has the potential to fundamentally change how people are selected for inclusion in clinical trials. In HD, people are currently selected based on relatively crude measures defined by having the mutant huntingtin gene, clinical presentation (e.g. motor, cognitive, behavioural), and the simplest of imaging measures (e.g. regional brain volume). This project will establish a framework for integrated information, which will combine existing biomarkers with computational methods to produce digital biomarkers, and reveal hidden information from large and varied datasets. This framework has the potential to impact on how clinical trials are designed, and hence will impact first on academic-industry collaboration (~3 years), and in the longer term (5-10 years) on the pharmaceutical industry itself.
iv) UK economy
The Alzheimer's Society has shown that dementia costs the UK economy £26.3bn per year, and estimates that this cost will double in the next 25 years. While rare, HD is a severely debilitating condition with an early age of onset (typically 35-50 years old) that has a devastating impact on patients' lives, and is usually fatal within 15-20 years from onset. The research from this Fellowship aims to accelerate the development and deployment of a disease modifying therapy for HD, which would significantly reduce the costs of care for people with HD. As such, in the longer term (> 10 years) this Fellowship has the potential to impact on the UK economy by reducing healthcare costs and enabling people to work.
i) Development of disease modifying therapies
The primary route to treating HD is through disease modifying therapies, which can slow the progression of the disease. The research conducted in this Fellowship aims to accelerate the development of such therapies, by selecting the most suitable group for testing a given therapy, and hence allowing clinical trials to become smaller (in terms of the number of participants) and shorter (in terms of the length of the trial). The focus of this project is on Phase 2 trial design, but it can naturally be extended to Phase 3, thus facilitating a fully computerised framework that can be used to optimise future trial design. With a current clinical trial for HD on-going, this framework has the potential to be validated using real trial data within the next 3-5 years. Once validated, it can then be applied to other neurological diseases, and ultimately any disease. This will impact on the clinical trials community (academic and industry), and people with HD.
ii) Clinical practice and patient care
The computational stratification tools produced by this Fellowship will help inform clinical decision making, by providing personalised patient information. These tools will help clinicians distinguish between different phenotypes, how these phenotypes are expressed together, and how a given phenotype is expected to progress in time. As part of this Fellowship, I will be trained at the UCL Huntington's Disease Multi-Disciplinary Team Clinic, which will help me ensure that the computational information can be used aid in treatment planning, therapy, and patient care. In the short-term (~3 years), this will impact clinicians in their views on modern computational methods and how they can be used in practice. In the longer term (5-10 years), the methods developed in this study could be used to impact on clinical decision making, and hence the care of people with HD.
iii) Pharmaceutical industry
The computational framework for clinical trial design proposed by this Fellowship has the potential to fundamentally change how people are selected for inclusion in clinical trials. In HD, people are currently selected based on relatively crude measures defined by having the mutant huntingtin gene, clinical presentation (e.g. motor, cognitive, behavioural), and the simplest of imaging measures (e.g. regional brain volume). This project will establish a framework for integrated information, which will combine existing biomarkers with computational methods to produce digital biomarkers, and reveal hidden information from large and varied datasets. This framework has the potential to impact on how clinical trials are designed, and hence will impact first on academic-industry collaboration (~3 years), and in the longer term (5-10 years) on the pharmaceutical industry itself.
iv) UK economy
The Alzheimer's Society has shown that dementia costs the UK economy £26.3bn per year, and estimates that this cost will double in the next 25 years. While rare, HD is a severely debilitating condition with an early age of onset (typically 35-50 years old) that has a devastating impact on patients' lives, and is usually fatal within 15-20 years from onset. The research from this Fellowship aims to accelerate the development and deployment of a disease modifying therapy for HD, which would significantly reduce the costs of care for people with HD. As such, in the longer term (> 10 years) this Fellowship has the potential to impact on the UK economy by reducing healthcare costs and enabling people to work.
People |
ORCID iD |
Peter Wijeratne (Principal Investigator / Fellow) |
Publications
Aksman LM
(2021)
pySuStaIn: a Python implementation of the Subtype and Stage Inference algorithm.
in SoftwareX
Eshaghi A
(2021)
Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data.
in Nature communications
Ghazaleh N
(2021)
Ranking the Predictive Power of Clinical and Biological Features Associated With Disease Progression in Huntington's Disease.
in Frontiers in neurology
Hadjicharalambous M
(2021)
From tumour perfusion to drug delivery and clinical translation of in silico cancer models.
in Methods (San Diego, Calif.)
Hithersay R
(2021)
Optimal age and outcome measures for Alzheimer's disease prevention trials in people with Down syndrome.
in Alzheimer's & dementia : the journal of the Alzheimer's Association
Oxtoby NP
(2021)
Sequence of clinical and neurodegeneration events in Parkinson's disease progression.
in Brain : a journal of neurology
Rodrigues FB
(2020)
Mutant huntingtin and neurofilament light have distinct longitudinal dynamics in Huntington's disease.
in Science translational medicine
Description | Learning personalised trajectories in Huntington's disease through computational models of disease progression |
Amount | £185,105 (GBP) |
Funding ID | A15920 |
Organisation | CHDI Foundation |
Sector | Charity/Non Profit |
Country | United States |
Start | 06/2021 |
End | 06/2023 |
Description | F. Hoffman-La Roche - UCL Computer Science collaboration |
Organisation | F. Hoffmann-La Roche AG |
Department | Roche Diagnostics |
Country | Global |
Sector | Private |
PI Contribution | I provide expertise and guidance on machine learning techniques to their Huntington's disease data science team. I also provide feedback on associated scientific publications (writing, content, analysis). |
Collaborator Contribution | My partners lead the analysis itself and prepare the research for dissemination. |
Impact | The collaboration is in its early stage but has produced research that has been presented at an international conference [1]. [1] Ghazaleh, Naghmeh, Houghton, Richard, Palermo, Giuseppe, Schobel, Scott, Wijeratne, Peter A, Long, Jeffery D. Discovering Dominant Features Associated with Disease Progression in Huntington's Disease: A Data-Driven Approach Using the Enroll-HD Database. Neurotherapeutics. 2019; 16(4):1361-1361 |
Start Year | 2019 |
Description | A-COMPS: Apply Competitively to STEM |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | Local |
Primary Audience | Schools |
Results and Impact | I am leading "A-COMPS: Apply competitively to STEM", an exciting new mentoring programme for secondary school students from disadvantaged backgrounds run by the UCL Department of Computer Science. A-COMPS is an 18-month programme aimed at A-level students across London, who are interested in computer science, technology and engineering and would like support with their journey into higher education. Mentees will be taught by leading experts in the field of computer science, given one-to-one academic and personal mentoring, work experience in our research labs, and will participate in expert-led workshops on UCAS applications, exploring their skills, strengths and STEM careers, writing personal statements and advice on student finance. |
Year(s) Of Engagement Activity | 2022 |
Description | Bloomsbury Festival |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Public/other audiences |
Results and Impact | Online panel discussion with PhD students and staff from the UCL Centre for Medical Image Computing, attended by members of the public and fellow academics as part of the Bloomsbury Festival. The discussion was lively and prompted future related activities, including a similar format planned for the Bluedot Festival. |
Year(s) Of Engagement Activity | 2020 |
URL | https://bloomsburyfestival.org.uk/2020vision/ |