Deliverable 2.3: Linkage and feature extraction from gut-brain, initial evaluation

The use case that this deliverable is associated to in the HEREDITARY project aims to replicate and expand earlier findings on gut-brain axis (GBA) interactions by analysing
the relationships between brain structure and function, gut microbiota composition, and metabolic parameters. This work leverages multimodal data from two datasets: the Healthy Brain Study (HBS), a cohort of 900 deeply phenotyped individuals aged 30-39 assessed longitudinally across one year, and the Mindset dataset, which includes over 200 psychiatric in-patients with behavioural, clinical, imaging, and microbiome data. The analyses integrate microbiome measures (relative abundance and diversity) with brain connectivity and structural data, using advanced Linked Independent Component Analysis (LICA) and other machine learning methods to investigate multimodal 
associations. These exploratory methods are meant to set the stage for the discovery of connections between the gut and the brain within HEREDITARY using data analysis 
techniques on multimodal data.

For this deliverable, preliminary analyses focused on a subsample of 234 healthy participants from the HBS cohort, with exclusions based on incomplete or poor-quality 
data. After processing, 173 participants were included in the analysis. The LICA model identified 11 independent components (ICs) that demonstrated significant contributions from both resting-state brain connectivity and microbiome relative abundance. These ICs revealed promising GBA patterns, with several components showing multivariate interactions rather than dominance by either modality. For instance, IC 26 highlighted covariance between the relative abundances of specific bacterial genera (e.g., Prevotella, Dialister) and brain networks, including the default mode network (DMN), salience network (SN), and executive control network (ECN). The findings support earlier studies and validate the applicability of LICA for identifying complex GBA interactions.

Future analyses within HEREDITARY will focus on exploring these GBA linkages in the full HBS cohort and conducting hypothesis-driven studies to investigate relationships 
between stress, diet, microbiome composition, brain connectivity, and eating behaviour. Additionally, the project will expand the integration of GBA data using emerging AI models such as foundation models in fMRI and digital pathology, with a particular focus on extending this type of technology to address gaps in microbiome data analysis. The results of these analyses are expected to advance understanding of gut-brain interactions, particularly in the context of stress, behaviour, and health. Initial findings are encouraging, as they replicate prior results from smaller samples and highlight the potential for multimodal machine learning to uncover novel insights into the GBA. Further work will aim to validate these findings, evaluate longitudinal changes, and explore their implications for understanding stress-related eating and other behaviours. A perspective article on the role of AI in multimodal GBA research is also planned for 2025 to address gaps identified in the literature.