Abstract

Alzheimer’s Disease (AD) is a neurodegenerative disorder contributing to 50 - 75% of all dementia cases. Amyloid plaques are accumulations of beta-amyloid proteins that aggregate between the nerve cells (neurons) in the brain of patients with AD and hence, their existence are salient pathological indicators for the disease. Previous research have found that an imbalance between production and clearance of the Amyloid-beta (A) and related A peptides is a very early, often initiating factor in AD [19]. Interestingly, a number of different plaque morphologies have been reported to correlate with different clinical features of AD. However, the relationship between these plaque features to the neurodegenerative process remains a central question in AD research.

Despite advancements in transcriptomics methods including high throughput single cell sequencing, for which analysis infrastructure is relatively well established, conventional transcriptomics methods commonly omit the spatial structure of gene expression within an underlying tissue. Recent technological developments in spatial transcriptomics (ST) allows the measurement of the expression of all genes in a tissue and retains spatial information with 100 micron resolution. This technical advancement has opened new opportunities to investigate the relationship between the amyloid morphological pattern and changes in topological gene expression, such as the genomic responses to the pathological features, the cell types and sub-types contributing to those responses, its dependency to the neighbouring tissue, etc. However, this new form of transcriptomics data has lead to unprecedented data analysis challenges, requiring the combination of two disparate data types: the expression level of several thousands of genes as well as their spatial information including general histology and pathological staining.

In this challenge, we have attempted to understand the relationship between amyloid plaque image and spatial transcriptomics patterns in Alzheimer’s disease mouse model using a range of machine learning methods. We have approached by first characterisation and extraction of the key features in spatial transcriptomic and A plaque stained images separately, followed by a brief exploration of the relationship between the two, and finally the comparison of machine learning models that predict plaque information from gene expression information.

Citation information

Data Study Group team. (2022, July 5). Data Study Group Final Report: UK Dementia Research Institute and DEMON Network. Zenodo. https://doi.org/10.5281/zenodo.6798982

Additional information

PIs: Abhirup Banerjee and Yeung-Yeung Leung