Data mining epidemiological relationships: integration of causal analysis with published evidence

Causal inference in epidemiology focuses on identifying the risk factors that cause disease. Established approaches focus on specific risk factors that may impact on specific diseases. However, the wealth of biomedical data that now exist enable us to assess the causal relationships between a broad network of risk factors and diseases. By considering a much wider network of such relationships we will establish the relative importance of different risk factors and the potential side-effects of interventions that target those risk factors. We will also integrate biological data (eg molecular pathways, drug targets) with causal relationships to enable us to understand the molecular mechanisms that lead to disease, and identify potential pharmaceutical and public health interventions. These data and relationships will be combined in a purpose-built “graph” database and methods will be developed to mine for novel causal risk factors and potential interventions.
The data that we collate for our research within this programme will have wide-reaching value to the research community. We will provide an open and accessible software platform for other researchers to search and use the various datasets we have integrated for their own research.

Background: The increasing availability of complex, high-dimensional epidemiological data necessitates innovative and scalable approaches to harness this power to address research questions of biomedical importance.
Aims: Motivated by the widespread adoption of Mendelian randomization and the opportunities to integrate multiple data sources for the triangulation of evidence in epidemiological research, this programme will develop and apply novel data mining approaches in integrative epidemiology. We will also develop and implement a software platform to enable research questions of major epidemiological importance to be addressed rapidly and at scale.
The programme will focus on (a) integration of cutting edge statistical methods under development in the MRC Integrative Epidemiology Unit (MRC-IEU) with extensive data in a graph database; (b) development of subgraph searching algorithms; and (c) identification of causal mechanistic pathways to disease. EpiGraphDB will be a resource of extensive value to the programme, the MRC-IEU and the wider research community.
Research plans: The programme will implement a data mining approach by developing a new graph database (EpiGraphDB) that will integrate cutting edge causal analysis evidence with comprehensive data on relationships between traits, risk factors, biomarkers, intervention targets and diseases. These data will originate from Mendelian randomization, genetic and observational correlation from epidemiological studies, relationships mined from the literature, and a wide array of bioinformatics sources describing molecular relationships. EpiGraphDB will enable aetiological hypotheses to be generated and explored.
Data sharing and health applications: The database, software and results generated by this programme will be made openly available to the wider scientific community for application to a range of potential health questions (eg identifying causal risk factors for disease, identifying side-effects of interventions, etc).