Statistical modelling of postprandial metabolomic and lipidomic profiles in individuals with varying glycaemic regulation using R Statistical Software

Abstract

Introduction: Dietary macronutrient composition, specifically the intake of meals enriched in
carbohydrates, fats, protein or fibre, triggers distinct postprandial metabolic responses but
whether these responses vary significantly based on an individual’s glycaemic status: type 1
diabetes (T1D), type 2 diabetes (T2D), or normoglycaemia (ND) remains an open question.
Background: Metabolomic and lipidomic profiling provide valuable insights into metabolic
alterations associated with diabetes. However, the main analytical challenge lies in the
analysis of large, high-dimensional datasets generated from complex experimental designs
involving multiple variables and repeated measures. Existing statistical model approaches
have identified limited diabetes-meal interactions, highlighting the need for complementary
machine learning (ML) methods capable of detecting group specific metabolic signatures in
complex data.
Aim: The aim of this project was to apply advanced statistical and predictive modelling
methods in R to characterise postprandial metabolomic and lipidomic responses to different
meal, and to investigate how dietary macronutrient composition modulates these responses in
individuals with varying glycaemic regulation.
Methods: This project involved analysis of metabolomic and lipidomic data from plasma
samples taken during a meal tolerance test where participants were given isocaloric meals
with varying macronutrient compositions, using linear mixed models (LMM) and machine
learning with SHapley additive exPlanations (SHAP) interpretation.
Results: LMM identified significant effects of time, diabetes and meal type on postprandial
metabolite and lipid levels. Random forest distinguished ND from T2D with over 80%
accuracy and predicted meal type with similar accuracy. Triangulation of SHAP and LMM
identified ten priority features, including fructose, plasmalogens and glycerolipids, showing
distinct postprandial response patterns, including chronic baseline elevation independent of
meal type and fat meal induced crossover responses between ND and T2D.
Conclusion: The combination of LMM and ML provided complementary diabetes-associated
postprandial signatures, with the fat-rich meal producing the greatest difference between ND
and T2D. These findings suggest that dietary macronutrient composition modulates
3
postprandial metabolism in a diabetes-dependent manner and may offer a dietary strategy for
reducing postprandial metabolic dysregulation in T2D.

Popular Abstract

How diabetes alters the body’s response to food
Every time we eat, our body starts a complex series of processes to digest food
and distribute energy to our cells. The blood fills with hundreds of different molecules such as
sugars, fats, amino acids, that rise and fall in the hours after eating. In healthy individuals, this
process is tightly regulated. However, in people with diabetes, this regulation is disrupted, and
the body responds to food differently. Diabetes affects millions of people worldwide and
includes conditions such as type 1 diabetes (T1D), characterized by an absence of insulin
production, and type 2 diabetes (T2D), in which the body becomes resistant to insulin.
This project investigated how different meal types affect blood metabolites and
lipids, and whether these responses differ between individuals with normal glucose
regulation, T1D and T2D. Participants consumed meals enriched in carbohydrates, fats,
proteins, or fibre. Blood samples were collected before and after eating to measure changes in
hundreds of small molecules involved in metabolism.
To analyse these complex data, we combined statistical models and machine
learning methods. Traditional statistical tools were first used to identify molecules that
changed over time depending on the meal type, time after eating, and diabetes status. Machine
learning models were then trained to determine which molecules contributed most strongly to
distinguishing between meals and the diabetes group.
Our results showed that the blood molecular profile carries strong signals about
both meal type and whether a person has diabetes. The machine learning model correctly
identified the consumed meal in nearly 81 % of cases and distinguished between diabetes
groups with 83 % accuracy. Several molecules, including fructose, certain fats and amino acid
derivatives, showed clear difference between healthy individuals and those with T2D.
Importantly, the metabolic response to food was not the same between ND and T2D,
suggesting that diabetes actively alters how the body handles food.
Taken together, these findings improve our understanding of how diabetes
reshapes postprandial metabolism and is a first step towards more personalized, food aware
approaches to managing metabolic health.

• author: Zaharova, Angelina
• supervisor: Peter Spégel
• organization: Applied Microbiology; Biotechnology; Biotechnology (MSc)
• course: KMBM01 20261
• year: 2026
• type: H2 - Master's Degree (Two Years)
• subject: Chemistry
• keywords: Diabetes, Lipidomics, Machine learning, Macronutrients, Metabolomics, Applied Microbiology
• language: English
• id: 9233557
• date added to LUP: 2026-06-10 13:14:11
• date last changed: 2026-06-10 13:14:11