Digging deep into data: early career researchers take on challenge of dementia

This World Alzheimer’s Month, early career researchers from across the UK and beyond have been hard at work harnessing data science and AI to solve real life problems in dementia and health research. Data Study Groups are ‘collaborative hackathon’ style events organised by The Alan Turing Institute, and the September 2021 event featured dementia challenges set by the UK DRI in collaboration with the DEMON Network. We found out more about the projects these data scientists embarked upon and the innovative solutions they have uncovered.

The Turing Data Study Groups are events that bring together data scientists and AI to tackle data driven research challenges set by organisations – the ‘Challenge Owners’. Importantly, the challenges use real-life data, so that the results of the Data Study Groups can have real-world impact. The Turing Data Study Groups aim to bring the power of computational approaches to bear on problems in research areas that may not have traditionally used data science and AI. Previous organisations who have taken part include partner universities like the University of Bristol and Siemens Global Centre of Competence Cities. Challenges posed range from how to automatically identify proteins from cryogenic electron microscopy images to how urban infrastructure can be optimised to reduce air pollution. Many of these Challenge Owners have continued to collaborate with The Alan Turing Institute and the researchers involved in the Data Study Groups.

Finding innovative ways to approach research is important to drive ground-breaking advances in this area and develop new technologies and treatments. The Turing Data Study Groups provided a fantastic opportunity to develop and test possible solutions to a variety of problems encountered in dementia research that required the computational skills of talented groups of researchers to solve. The challenges posed are described below:

Modelling amyloid beta protein plaque formation in Alzheimer’s disease

Prof Bart De Strooper (UK DRI Director) and Dr Mark Fiers (VIB-KU Leuven)

Amyloid beta protein plaques are a hallmark of Alzheimer’s disease and are thought to be associated with disease progression. The plaques exist in different shapes; however, it isn’t known whether or not these shapes have different effects on the surrounding cells in the brain. Establishing this link with image analysis and machine learning methods could help guide potential treatments that target amyloid plaques, maximising their benefit by focussing on the most harmful plaques.

Predicting functional relationships between DNA sequence and epigenetic state: Can computational models consider important genomic variants? 

Dr Nathan Skene (UK DRI at Imperial)

Current deep learning models can determine where DNA will bind regulatory proteins and its likely epigenetic state at that region. However, these models are based on a single reference genome, so it isn’t known how the accuracy of the models is affected by variants that exist in different genomes. Validating that these models can take into account genetic variants will allow them to predict how such variants associated with Alzheimer’s contribute to the disease mechanism through epigenetics. This could help identify important molecular and cellular steps in the disease as potential therapeutic targets.

As Dr Skene explained further:

“Our group are trying to establish which of the millions of genetic variants common in humans affect neurodegenerative diseases and how. This is impractical to study experimentally, so we’ve proposed a project to find the genetic variant effects computationally using machine learning.”

Using machine learning to improve sleep habits in people living with dementia

Dr Eyal Soreq, Prof Payam Barnaghi and Prof David Sharp (UK DRI at CR&T)

Sleep disorders can be a problem for people with dementia. By using data from multiple sensors that recorded the sleep patterns and conditions of people living with dementia, the project aims to identify how sleep metrics are affected by changes in different conditions, and determine if models can suggest condition changes that could improve sleep quality.