MRC DiMeN Doctoral Training Partnership: Quantifying and incorporating information bias in modern life science

Modern life sciences are increasingly driven by next-generation sequencing and whole genome/transcriptome/proteome/metabolome wide-association studies. Such ‘modern life science’ associates large-scale datasets with phenotypes or diseases of interest to discover their underlying biology to be used in medicine. With such discovery-based science a lot of progress can be made, however, there is a large dependency for interpretation on currently known information. This information bias generates self-fulfilling prophecies, biased research directions and interpretations. We know that the level of information is heavily biased - for example, because, certain genes, diseases or species have been hotspots of research (https://www.nature.com/articles/d41586-017-07291-9). Even worse, such study bias reinforces information bias further and it is often unclear what the reliability of the initial information was in the first place. Here we propose a statistical approach that incorporates information bias in both quantity and quality to objectively distil biology of interest and to flag up important currently unknown biology.

The techniques used to achieve this will involve machine learning (principle component and cluster analysis), evolutionary genomics, data imputation and reliability weighting (as in meta-analysis). Moreover, the student will help generate a user-friendly GUI for the community. For proof of principle we will utilise one database, Flybase (Drosophila melanogaster), that is very well annotated and has the largest quantity of in depth information not available for other organisms. The methods generated will however be applicable to any database, including human databases. In addition, the use of flies will allow the student to put the software generated to the test using two large experimental datasets that we have generated in my lab: experimental GWAS in flies for longevity and neurodegenerative disease. Candidate genes identified by the student can subsequently be tested for causality using functional genetics of flies.

Modern life science generates a tremendous amount of data that is close to incomprehensible except to study the top associations with the outcome (e.g. disease) of interest. Currently such top associations, or ‘hits’, are compared to a database of current knowledge on the gene level. Such data mining can be rewarding, but also imposes strong information bias. The beauty of modern life science is that it has the potential to be unbiased and can be highly informative about mechanisms, especially in disease or for complex phenotypes such as ageing. The project we propose is therefore exceptionally timely. Biological interpretation from large scale, often costly, studies that are currently generated is impeded by our current inability to assess current knowledge objectively, intuitively and in an integrated fashion.

We seek a hard-working passionate student with a quantitative mind-set and an interest in the biology of disease and ageing. Several educational backgrounds are suitable for this project and include, but are not exclusive, to bioinformatics, computational science and biomedical science. The techniques that are listed can all be learned within the PhD and no prior experience with these is essential. The broad scientific expertise and commitment of the supervisors ensures ample support and a vibrant environment for the student.

Links:
http://simons-lab.group.shef.ac.uk
http://toni-gossmann.staff.shef.ac.uk