Metabolic Health

Type 2 Diabetes (T2D) is not a single disease, but a complex metabolic condition shaped by the interplay between internal biology and lifelong environmental exposures (the exposome). While standard clinical tests offer a snapshot of glucose levels, they fail to capture how chronic environmental stressors, lifestyle factors, and molecular changes combine over time to drive metabolic decline.

Building on our published physiological framework that links chronic stress (the HPA axis) to glucose dysregulation in the UK Biobank [1], this project takes the next step. We aim to combine mechanistic physiology with modern machine learning, integrating large-scale multi-omics and exposomics data to build predictive models of personalized metabolic trajectories.

Objectives

1. Integrate high-dimensional exposome data (chronic stress indicators, sleep patterns, diet, socio-environmental factors) with multi-omics profiles (genomics, metabolomics, proteomics) from large-scale cohorts like the UK Biobank to discover multi-modal risk signatures.
2. Extend our existing mechanistic HPA–metabolic framework into a hybrid AI model (such as physics-informed neural networks or biologically constrained learning) that predicts individual glucose dysregulation paths over time.
3. Apply multi-view clustering and explainable AI techniques to uncover distinct T2D sub-phenotypes, mapping out how specific environmental stressors interact with unique biological vulnerabilities.

Starting datasets

• NHANES — https://wwwn.cdc.gov/nchs/nhanes/Default.aspx — identify which environmental stressors (sleep, diets, toxins) most strongly correlate with T2D clinical trajectories.
• iHMP Prediabetes — https://www.hmpdacc.org/ihmp/ — continuous glucose monitoring, dense multi-omics and stress proxies (HRV/sleep) to constrain the mechanistic HPA–metabolic ODEs.
• All of Us — https://www.researchallofus.org/institutional-agreements/ — register via AmsterdamUMC or VU.

References

1. Getsadze, R., et al. "Mechanistic Modeling of Stress-Induced Glucose Dysregulation: Calibrating a Coupled HPA-Metabolic Framework to the UK Biobank" (2026).
2. Zhou, W., et al. "Longitudinal multi-omics of host-microbe dynamics in prediabetes." Nature 569.7758 (2019): 663–671.
3. Ahlqvist, E., et al. "Novel subgroups of adult-onset diabetes and their association with outcomes." The Lancet Diabetes & Endocrinology 6.5 (2018): 361–369.

Work environment

The student will join a collaborative team across the Informatics Institute (IvI) and Amsterdam UMC, working closely with experts in AI, complex systems, and clinical endocrinology. You will gain hands-on experience with population-scale health data, multi-modal integration, and hybrid machine learning models applied directly to personalized healthcare.

Under the Developmental Origins of Health and Disease (DOHaD) paradigm, the first 1,000 days of life represent a sensitive window shaping long-term susceptibility to metabolic diseases like type 2 diabetes. A primary driver of this mechanism is the gut microbiome and early-life exposome (e.g. diet, antibiotics, delivery mode), which modulate physiological metabolism via systemic metabolites. To capture these complex interactions, recent work couples Bayesian Directed Acyclic Graphs (DAGs) with Physiologically-Based Kinetic (PBK) models [1].

This project focuses on building, validating, and coupling the causal DAG backbone [2]. You will leverage literature mining and clinical cohort data (stool multi-omics, exposome factors, and blood biomarkers) to infer causal network topology. Using Bayesian structure learning and sensitivity analysis, you will identify key exposome–microbiome–health pathways and integrate your validated DAG into the pre-existing PBK framework.

Objectives

1. Data scraping & integration: literature and internal datasets (exposome, microbiome, biomarkers).
2. Causal DAG construction: apply Bayesian structure learning and constraint-based algorithms to map causal topology.
3. Model validation: conditional independence testing and sensitivity analysis on key pathways.
4. PBK coupling: integrate the validated causal DAG framework with an existing PBK model.

References

1. Ratier et al. (2024). https://doi.org/10.1016/j.envint.2024.108621
2. Ly et al. (2026). The 6th International Symposium on Complex Systems, ISCS 2026.

Expectations

Good programming skills (Python).

Work environment

The student will join a diverse team of researchers from the Informatics Institute (IvI) and the MetaHealth project (NWO). You will contribute to a computational model aimed at predicting early-life metabolic programming and informing targeted early-life interventions.