Published September 23, 2026 | Version v1

Deliverable 2.6: Clinical value assessment of gut-brain integration

  • 1. ROR icon Radboud University Medical Center
  • 2. ROR icon University of Turin

Description

This deliverable reports the initial clinical value assessment of gut–brain integration within Work Package 2 (D2.6), building directly on the multimodal integration and predictive modelling developed in Deliverable 2.4. D2.4 established that a supervised Linked Independent Component Analysis (SuperBigFLICA) applied to resting-state fMRI networks, gut microbiota relative abundance, and behavioural/physiological phenotypes from the Healthy Brain Study (HBS) yields a multivariate gut–brain component (Component 93) that significantly predicts valuation and choice of unhealthy foods in an previously unused behavioural task (Food Auction Task; FAT). This analysis forms the foundation of the extraction of clinical value. The gut–brain phenotype trained on physiological reconstruction and a limited set of behavioural and physiological targets has the potential to generalise to an unseen, disease-relevant outcome. This is the central requirement for any biomarker intended to support diagnosis, prediction, or treatment stratification.

We complement this evidence with independently published complimentary longitudinal evidence for the general principle that multimodal gut–brain integration carries predictive, clinically relevant information. Using longitudinal data from the same 702 participants of the Healthy Brain Study and random-intercept cross-lagged panel models, a recent peer reviewed review by our group (López-Sánchez et al., 2026) shows that salience- and executive-network connectivity prospectively predicts subsequent gut microbiota diversity, over and above stable between-person associations, with physical activity acting as a plastic driver of these brain-network changes. We summarise this evidence and outline its relevance for identifying candidate clinical targets and monitoring windows for gut–brain interventions in obesity.

Finally, we outline ongoing work, led by Agata Polejowska, on an automated, interpretable machine learning framework for microbiome analysis (“Optimising machine learning for interpretable microbiome analysis: an automated modelling framework”), which is intended to provide the model-selection and interpretability infrastructure required to translate exploratory multivariate components into clinically usable prediction models.

Taken together, these three lines, internal predictive validation, external longitudinal replication of brain-to-gut predictive pathways, and emerging interpretable ML infrastructure, support the feasibility of extracting disease-relevant associations of clinical value from gut–brain integration, consistent with the M33 objective of D2.6 for initial results from the HBS sample.

Files

HEREDITARY_D2.6.pdf

Files (2.0 MB)

Name Size
md5:9e28749a9b0056f986ad8140c42b508c
2.0 MB Preview Download

Additional details

Funding

European Commission
HEREDITARY - HetERogeneous sEmantic Data integratIon for the guT-bRain interplaY 101137074