A fully-quantifiable aetiopathogenic model for Parkinson’s and Overlap and Related Diseases to identify targets for personalized intervention

Lead Supervisor
Prof Steven Gilmour
Professor of Statistics
Department of Mathematics, Faculty of Natural & Mathematical Sciences
King’s College London
steven.gilmour@kcl.ac.uk

Co-supervisor
Dr Sylvia Dobbs, Dr John Dobbs, Dr André Charlett

Project Details

The PhD student would hit the ground running in consolidating the knowledge attained from measurements, clinical and on biological samples, made in the existing and expanding cohort of 200 participants (Parkinson’s disease (PD) probands, their spouses/partners, and controls with and without a family history), all being followed-up longitudinally, some already reaching 5 years. Potential to create leadership lies in our integrated approach to building a model to encompass PD as a systemic disorder, the submerged ‘iceberg’ (pre-presentation states, attenuated/partial PD manifestations), co-morbidities and overlap/related disorders. Other groups have studied one or two of aspects (phenome, quantitative metagenomics, immune modalities, metabolome and human gene variants), and ignore the iceberg and wider disease spectrum. Moreover, here, objective continuous measures of the facets of PD are used wherever practicable in describing the phenome.

The student will be involved in systematic development of a fully-quantifiable aetio-pathogenic model. The incremental plan has 4 steps:- 1. Appropriate dimension reduction within functional data-sets (e.g. principal components, two-way orthogonal partial least squares analyses) to identify essential and redundant components within each set. 2. Classification algorithms (multinomial logistic regression, classification and regression tree, random forests) to predict, from the essential data, membership signatures of participant categories and, then, position by measures of disease facets (‘distance-down-the-pathway’/severity) and co-morbidity. 3. Quantitative integrated classification model, derived by formal combination of data-set classifiers from cross-sectional and longitudinal data (novel, expert-guided, machine learning ensemble methods), with description of dynamic transition between positions in disease spectrum with and without intervention (multi-state hidden Markov model). 4. Multivariate functional time-series models will be developed to understand factors associated with evolution (enabling design of pivotal Randomised Controlled Trials (RCTs)), with circumscribed cross-referencing of position within aetio-pathogenic model against functional neuroimaging (dopamine-uptake, microglial activation).

Development of novel statistical methodology is required for several aspects of design and analysis:- 1. Design issues for functional data have hardly been studied and even the sample size calculations require new methodology to be developed. 2. Data collection will be sequential and adaptive, combining methods of adaptive design with outputs of learning algorithms. 3. Algorithms will be guided by expert knowledge (Drs John & Sylvia Dobbs with Medical Consultants from relevant disciplines), so that they identify clinically relevant variables and scenarios rather than just describing differences between participants. 4. Functional time series models will be developed for use on clinical data (e.g. emphasis on identifying and modelling smooth but complex trends, as well as sudden jumps). 5. Use of a complex multivariate functional time series model to inform the design of relevant RCTs, with particular attention to appropriate objective outcome selection.
Timeliness for further development lies in advancement of healthcare technologies research being a global priority. The scientific community is moving towards our main hypotheses. Pharma is currently rejecting global score outcome criteria, moving towards objective assessment of disease facets. Media and public are aware of the potential influence of the gut brain axis in neuropsychiatric disease and the focus on the gut microbiome.

The Host-Microbiome Interaction Research Group, in the Institute of Pharmaceutical Science at King’s College London (https://www.kcl.ac.uk/lsm/research/divisions/ips/research/major-research-themes/hmi-cpt/index) (core group Sylvia & R John Dobbs, clinical neuropharmacology; André Charlett, statistics and modelling) have an established collaboration with the Department of Mathematics, King’s College London (Steven Gilmour, Professor of Statistics and Head of Department).

The student will be involved in aspects of systematic development of a fully-quantifiable aetio-pathogenic model for Parkinson’s and overlap diseases. It will consolidate the knowledge attained from measurements, clinical and on biological samples, made in the existing and expanding cohorts. This development has 4 steps:- 1. Appropriate dimension reduction within functional data-sets 2. Classification algorithms. 3. Quantitative integrated classification model, derived by formal combination of data-set classifiers from cross-sectional and longitudinal data. 4. Multivariate functional time-series models will be developed to understand factors associated with evolution. Development of novel statistical methodology is required for several aspects of design and analysis.

Datasets

Gut microbiome drivers and mediators in the aetiopathogenesis of Parkinson’s disease and co-morbidities
Study Reference: HR-15/16-2810

Keywords

Statistical modelling, Parkinson’s disease, phenome, immunome, exposome, human genetics, alimentary metabolome and microbiome