Nutrition, genetics and the ageing brain: data analysis with application of artificial intelligence approaches to investigate brain health outcomes in older adults

Abstract

Dementia, characterised by a decline in memory, thinking, and decision-making abilities, is a debilitating condition that affects millions of older adults worldwide. The complexity of dementia arises from its multifactorial nature, with a combination of genetic, environmental, and lifestyle factors known to contribute to its development and progression over time. This thesis employs a data-driven approach, using advanced machine learning techniques and genome-wide association study (GWAS) approaches, to investigate the complex relationship of cognitive health with genetics and nutrition in older adults. A comprehensive systematic review and meta-analysis of randomised controlled trials assessed the effects of individual nutrients and dietary patterns on cognitive outcomes. The analysis revealed that supplementation for at least one year with B-vitamins, and to a lesser extent vitamin D, improved memory and global cognitive function, respectively.

To investigate genetic variants associated with cognitive dysfunction, a GWAS was conducted using data from the Trinity-Ulster-Department of Agriculture (TUDA) cohort. The TUDA study of over 5000 older adults allowed for a comprehensive analysis and in-depth investigation of the determinants of cognitive health, thus enabling a range of relevant genetic, along with sociodemographic, clinical, and lifestyle factors to be identified. A genome-wide analysis of this richly phenotyped cohort identified two significant single nucleotide polymorphisms (SNPs) - rs429358 (Apolipoprotein E gene and rs3771791 (hexokinase 2 gene) associated with cognitive dysfunction. In addition, three advanced machine learning models were developed to predict cognitive dysfunction from biological, health, and nutritional data, as well as identifying genome-wide SNPs. The best performing model - random forest –showed that plasma vitamin B6 and plasma homocysteine were critical biomarkers predictive of cognitive dysfunction. Further analysis explored gene-nutrient interactions and validated an important interaction of the ApoE ε4 allele with B-vitamin deficiency. Low status of vitamin B12, B6, and riboflavin, and elevated homocysteine concentrations, were associated with an increased risk of cognitive dysfunction, irrespective of ApoE status. Further exploration into the interactions between the ApoE ε4 allele and low B-vitamin status uncovered a novel interaction of ApoE ε4 and vitamin B12 deficiency (as measured by holotranscobalamin, the active form of vitamin B12), which was associated with an increased risk of cognitive dysfunction.

Analysis of the ‘TUDA5+’ longitudinal study was carried out using machine learning techniques, to identify key factors at baseline predictive of cognitive decline over a five-year follow-up period. In the TUDA5+ study, almost 20% of TUDA participants (n = 953) were resampled after five years following initial sampling for the full range of biomarkers, health and lifestyle measures. Additionally, genome-wide SNPs associated with cognitive dysfunction in an earlier analysis, were added to this rich longitudinal dataset. The best performing machine learning model identified cardiovascular factors, such as dizziness and self-reported family history of heart disease, along with low status of B-vitamins (folate and vitamin B12) and vitamin D as predictors of cognitive decline over time. The results emphasise the importance of using diverse data and highlight that multiple factors can each play a role and interact in contributing to cognitive decline, underscoring the multifactorial nature of cognitive health. Despite the established association between ApoE ε4 and cognitive dysfunction, modifiable factors, particularly B-vitamin status, were identified as important predictors of cognitive function and rate of cognitive decline.

In conclusion, this thesis underscores the importance of applying a multifactorial approach from different disciplines to better understand cognitive health, emphasising the interplay between nutrition and genetics, particularly between B-vitamin status and ApoE ε4.The findings also provide insights into the multitude of factors at play and the potential preventive strategies and stress the need for personalised dietary interventions for at-risk populations. By combining artificial intelligence and genetic insights to a phenotypically rich cohort, this research contributes to the growing body of evidence for the prevention and management of cognitive dysfunction in older adults.

Thesis is embargoed until 6th June 2027

Date of Award	Jun 2025
Original language	English
Supervisor	Catherine Hughes (Supervisor), Leane Hoey (Supervisor), Helene McNulty (Supervisor) & Mary Ward (Supervisor)