Data mining and bioinformatics cross-cutting theme

The investigation of how molecules influence health and disease (molecular epidemiology) has benefited from rapid development of methods to collect molecular data. This has created many challenges in how to handle, analyse and interpret the resulting “big data”. For example, we hold data on up to 2.5 million genetic variants and 450,000 epigenetic measures collected from thousands of samples. There is also a wealth of publicly available data that has the potential to give us further insights into disease processes when combined with the new data we collect.
The data mining and bioinformatics theme aims to: (a) develop and apply methods to combine data from a wide range of sources; (b) use publicly available data to filter and prioritise our results; (c) develop and apply methods to summarise data to reduce the computing power required; (d) “Mine” the resulting datasets to identify relationships between molecular, lifestyle and disease measures; (e) Train new and existing researchers in the methods required to work with “big data”.
This theme will build on core computational expertise in the key areas of molecular epidemiology, maximising the potential discoveries from the wealth of data generated by the programmes within the Unit.

A core feature of the Unit will be the generation, storage and analysis of large volumes of epidemiological and “omics” (genomic, epigenomic, metabolomic and metagenomic) data from a range of cohort studies. The Data Mining and Bioinformatics theme will be responsible for the development of novel tools and strategies to access, integrate, merge, visualise and analyse the multi-dimensional datasets within and across the various studies involved in the Unit Programmes. In particular the application of data mining and machine learning algorithms will allow hypothesis-free approaches to data interrogation (Programme 1), which can be followed by hypothesis testing in the other Programmes. Bioinformatics will also underpin the genetic, epigenetic and omics association studies that will be undertaken in all of the Programmes.
Objectives and plans:
Integration of diverse data types, including large-scale “omics” datasets: integration of very high-dimensional omics data to facilitate data mining and specific hypothesis-based analyses.
Filtering and annotation of omics association results: integration of internally generated data with public data from projects such as ENCODE, the NIH roadmap epigenomics project and others to prioritise and filter results.
Developing approaches for combining variables in omics data: collapsing and dimensionality-reduction approaches will be developed and applied to reduce the computational burden of association analysis between high-dimensional omics datasets.
Development of methods for hypothesis-free mining of complex data: using allele-score based approaches to de novo identification of potential causal relationships across a wide range of phenotypes/outcomes.
Training in bioinformatics and data-mining: specific training will be developed to build in-house skills in bioinformatics and data mining methodology in the context of high-dimensional data.
This theme will further develop the existing core expertise in bioinformatics and data mining to enable Unit researchers to work effectively with the data generated by the Unit and collaborators. The new methods developed will be of broad applicability in the field of molecular epidemiology, and the Unit will train a new cadre of researchers with essential skills in medical bioinformatics.