AMS Professorship award for Professor Dennis Wang, Imperial College London

Cardiovascular conditions such as hypertension are observed in ~44% of dementia cases.

And although specific diseases like pulmonary hypertension and vascular dementia may seem unrelated, transcriptional signals for the dysregulation of endothelial growth factors and inflammatory markers have been found in the patient cohorts of each disease.

By identifying molecular patterns in specific disease cohorts and then searching for their co-occurrence in ageing populations, we can not only identify early markers of multi-morbidities but also explain how a patient’s health trajectory changes over time leading to more serious conditions.

The lack of definitive diagnosis and measurable symptoms currently limit the detection and documentation of multi-morbidities in cohort studies.

Nevertheless, I have access to cardiovascular disease cohorts like the National Pulmonary Arterial Hypertension Study (n=1.2k), dementia cohorts CFAS, ADNI and ROSMAP (n=11k), multi-ethnic cohorts BiB, GUSTO and SPRESTO cohorts (n=20k), and ageing cohorts SAIL (n=3M), WISIC (n=2.3M) and UK Biobank (n=500k), where the phenotypes are becoming increasingly longitudinal and multi-omic, therefore, ripe for finding new measures of multi-morbidities.

During my professorship, I will investigate common molecular classifications for tightly phenotyped diseases (e.g. pulmonary hypertension) and a broad array of conditions relating to dementia.

For instance, signatures found in both development and disease, like tissue remodelling via endothelial cells, may exist in ageing populations. I will overcome the data integration and machine learning (ML) challenges which will arise in four stages: Years 1-2. Obtain ethics approval and aggregate data from disease and population cohorts.

Reduce missing -omics data and harmonize clinical variables across datasets.

Build an integrative platform with partners Sage Bionetworks and Turing Institute to manage and share the linked datasets. Years 2-3.

Extend unsupervised ML from Sokratis et al. (2021) to genome, proteome and methylome data in order to identify molecular signatures of pulmonary hypertension and dementia.

Visualise multi-omic profiles of each patient on the data platform and train clinical researchers at the host institution to organise clinical data. Years 2-4.

Model “many-to-many” associations between signatures and longitudinal clinical measures in disease (PAH, CFAS, ROSMAP) and multi-ethnic population cohorts (BiB, GUSTO, SPRESTO). Identify multi-morbidities in each cohort using molecular signatures and associated clinical measures. Years 3-5.

Train and apply predictive ML using multi-task Gaussian Processes (Wang et al. eLife 2020) to forecast the development of hypertension and dementia symptoms in large population cohorts (SAIL, UK Biobank, WISIC). Follow-up and verify onset of disease. AMS support will build capacity and resilience into my research programme.

In Years 1-2 it will enable me appoint a dedicated data scientist to harmonize variables using OMOP data standards and sequence additional samples to reduce -omics data missingness. It will also provide training support to allow clinicians to interact with the linked datasets.

In years 2 to 3, it will provide software engineering to scale the algorithms to run on multiple cohorts (n>1M) and multi-omic data types, leveraging my host institution’s resources and current grants (Wellcome, EPSRC) which currently support analysis of only a small longitudinal cohort (n=1000) with methylation and RNA data.