Data mining epidemiological relationships

We aim to develop and use cutting edge data mining tools to identify risk factors that cause common diseases and potential drug targets that could prevent or treat these diseases. Methods developed within the MRC Integrative Epidemiology Unit use genetic data to help identify lifestyle risk factors that could be modified to reduce the risk or impact of disease, and can also identify potential drug targets. This programme is developing tools and databases to automate this type of analysis and apply it to large-scale population datasets to help us discover new ways to prevent and treat disease. We are also combining the evidence from these analyses with other types of biomedical information in a “knowledge graph” to enable us to investigate the mechanisms underlying disease, identify new targets for treatment or prevention, predict side effects of drugs and identify opportunities to repurpose existing drugs for other diseases. The methods, software and knowledge graph we are developing are made openly available to the research community to maximise their potential to improve population health.

Background: Mendelian randomization (MR) is typically used to address specific causal hypotheses. Our MR-Base platform and OpenGWAS database now enable more systematic MR analyses of causal relationships between many traits and diseases, whilst our EpiGraphDB knowledge graph integrates these results with other biomedical evidence. Despite successes in identifying intervention targets and repurposing opportunities, such systematic MR analyses still face unsolved challenges in their interpretation and integration with other knowledge.
Aims: We aim to further advance approaches for systematically generating and integrating evidence to identify and prioritize intervention targets for disease prevention and treatment, and make these approaches and data resources widely accessible.
Objectives: (1) Developing and applying knowledge graphs (KGs) to generate hypotheses: we will use EpiGraphDB (and other KGs) for systematic analysis of specific disease outcomes, explore the use of graph embedding/link prediction methods to improve KGs and identify novel hypotheses, and develop natural language KG query interfaces to broaden their applicability. (2) Automating triangulation and evidence synthesis: we will develop new approaches to extracting evidence from the literature, websites and clinical trials databases. We will then systematically integrate this with evidence from MR and observational studies (including target trial emulation) and explore approaches to automating triangulation and synthesis of evidence for intervention targets. (3) Identifying and prioritizing intervention targets: we will use transcriptomic signatures to identify off-target side effects, strengthen the evidence for drug targets by integrating molecular QTL (molQTL) across traits and tissues with literature, coding mutations (including autozygous loss of function mutations) and animal knockouts, and implement approaches for identifying interactions. We will further develop trans-ancestry MR for prediction of cross-population generalisability of both pharmaceutical and non-pharmaceutical interventions. (4) New software and data resources: we will develop new open data and software resources based on IEU methodological innovations. We will enhance OpenGWAS by integrating non-European GWAS datasets to support multi-ancestry MR, implementing variance GWAS to identify potential interactions and improve automated phenotype curation and clustering. We will also implement a new curated molQTL catalogue to support drug-target MR.
Importance: This programme will develop and apply systematic approaches to prioritise and validate causal hypotheses, linking methodology developed in the unit with applied epidemiological research. Implementing these approaches in open software/data resources and applying them to emerging datasets will yield new discoveries to improve population health.