## Identifying and harmonising datasets We will collate datasets with data on medical conditions with at least one of: - cognition - genetics - neuroimaging This will include population-based datasets (eg [UK biobank][1], [studies used in the COSMIC project][2]), datasets examining cognition (eg [Dementia Platforms UK][3], [GAAIN][4]), and genetic datasets, (eg [PROTECT available through DPUK][5]) ## Harmonising cognitive scores - the Cross-Cohort Cognitive Composite We will use a method developed by the [Adaptive Brain lab][6] lab that has been applied previously to harmonise 4 disparate datasets with partially overlapping cognitive tests. The outcome is an estimated generic cognitive score for all datasets. For example, if two datasets had collected cognitive test A, the first having also collected test B, and the second test C, the available data from A, B and C could be used to generate the Cross-Cohort Cognitive Composite score. The first step is to harmonise data using a k-nearest neighbours imputation. Secondly, average scores are calculated for cognitive domains (eg memory, executive function). These scores are combined to form the Cross-Cohort Composite Cognitive Score. [See the poster describing the method here][7] ## Standardised neuroimaging processing We will establish standard preprocessing pipelines for the neuroimaging data to support predictive models. There is extensive experience within the DEMON network of applying neuroimaging pipelines to large datasets. Through the development phase we will build on previous work designing and implementing automated analysis pipelines for different MRI sequences that can accommodate high volumes of scans with minimal operator input and generalisable to different MRI manufacturers. Harmonised and standard pipelines are emerging for processing a range of MRI sequences through the [Organisation for Human Brain Mapping][8], such as COMBAT (Pomponio et al., 2020) and COMBAT-GAN (Fortin et al., 2018) for minimising variability between scanners. These pipelines draw on those used in the UK biobank dataset. We will apply these preprocessing steps where possible. [1]: https://www.ukbiobank.ac.uk/ [2]: https://cheba.unsw.edu.au/consortia/cosmic/studies [3]: https://portal.dementiasplatform.uk/ [4]: https://www.gaaindata.org/partners/online.html [5]: https://portal.dementiasplatform.uk/CohortDirectory/Item?fingerPrintID=PROTECT [6]: https://www.abg.psychol.cam.ac.uk/ [7]: https://mfr.de-1.osf.io/render?url=https://osf.io/9ar32/?direct%26mode=render%26action=download%26mode=render [8]: http://www.humanbrainmapping.org