2018-06-12 Project Team Meeting - Agenda & Minutes

Minutes

Data repository opportunities

>Nick Lynch provided his view on our approach to 'data sharing with external collaborators' based on his experience from working on information policy at AstraZeneca, IMI grants e.g. OpenPHACTS and recent general support for pharma in data management:

• Focus on common parts which can be shared easily and assess whether they provide enough value to get started
• Information which can be shared easily is seen as information which can be put in a publication or public database e.g. PDB and will make it through the standard publication process
• Sharing of metadata on assays as well as the data itself are equally important and are the basis for long term exploitation
• Sharing of metadata can perhaps be done without the legalities?
• Investing in establishing common terminology is critical in adding long term value
• Minimum Information Standards (see: https://www.ncbi.nlm.nih.gov/pubmed/21878981 and http://isa-tools.org/) and FAIR principles (https://www.nature.com/articles/sdata201618) are two patterns/initiatives to base this on and adhere to
• An example where 20k pre-clinical AstraZeneca assays were agreed to a common standard was used to illustrate a potential approach - this is not dissimilar to our proposed approach and involves understanding the purpose of the individual assay and what will be done with the data generated in terms of serving the decision-making e.g. TPP. Next step is to focus on up to 10 properties e.g. technology used, experiemntal measurement, variables, identifiers used in the public domain e.g. gene ontology, etc.

>The sense is that we are potentially close to being able to establish this common language and do this cross-comparison for some of the assays with low variability but the 'devil will be in the detail'. One likely criterion which will present challenges is 'buffer condition' as many assays will be buffer and concentration dependent. If not able to compare directly will need to use simple R/A/G categories to compare.

>Our discussion highlighted that we need a clearer idea of what we are going to do with the data, what we are going to use the data for and how we are going to analyse it e.g. will R/A/G descriptors may allow for comparison but  will they allow for building of predictive models?