Computer Model Predicts How Well GLP-1 Peptide Drugs Will Lower Blood Sugar From Lab Data Alone | Research

The takeaway

A systems pharmacology model accurately predicted the long-term blood sugar-lowering effects of GLP-1 and dual GLP-1/glucagon peptide drugs using only early lab and pharmacokinetic data.

5.9% prediction error
The model predicted fasting plasma glucose responses to GLP-1 peptide drugs with high accuracy using only in vitro and pharmacokinetic data

What the researchers found

The combined 4GI-HbA1c systems model successfully predicted the glucose and HbA1c-lowering effects of both liraglutide (a GLP-1 agonist) and cotadutide (a dual GLP-1/glucagon agonist) using only in vitro potency data and pharmacokinetic information. The model was validated against continuous glucose monitoring data from Phase 2a studies and achieved prediction errors of 5.9% for fasting plasma glucose and 13% for HbA1c.

Critically, the model was used prospectively during cotadutide's clinical development to predict 26-week glucose and HbA1c outcomes of a Phase 2b study before the study was initiated — and retrospective analysis confirmed the predictions were adequate.

Why it matters

Developing peptide drugs is expensive and slow. This model allows pharmaceutical companies to predict how a new GLP-1 or dual agonist peptide drug will perform in clinical trials using only early-stage lab data, potentially saving years of development time and significant costs. The ability to forecast long-term HbA1c outcomes from in vitro potency data alone could accelerate the selection of the most promising peptide drug candidates before committing to large, expensive clinical trials.

The numbers in context

RMSPE 5.9% for fasting glucose prediction · RMSPE 13% for HbA1c prediction · 26-week outcomes predicted prospectively · 2 peptide drugs validated (liraglutide, cotadutide) · Phase 2a CGM data used for calibration

How the study worked

The researchers extended their previously developed 4GI glucose homeostasis systems model by coupling it with an existing integrated glucose-red blood cell-HbA1c (IGRH) model. The model translates in vitro potency and pharmacokinetic data into predicted 24-hour glucose profiles, then uses these to forecast HbA1c changes over time. Validation used continuous glucose monitoring data from Phase 2a clinical trials of liraglutide and cotadutide.

Who was studied

Systems pharmacology modeling study; validated against clinical trial data from Type 2 diabetes patients in Phase 2 studies of liraglutide and cotadutide

What this study cannot tell us

The model was validated on only two peptide drugs (liraglutide and cotadutide), and broader validation across more GLP-1 agonists and dual agonists would strengthen confidence. The 13% prediction error for HbA1c is moderate and may not be sufficient for regulatory decision-making in all contexts. The model focuses on glucose control and does not predict other important outcomes like weight loss, cardiovascular effects, or safety profiles. Model calibration still required some clinical CGM data.

How to read the evidence

This is a computational pharmacology study that developed and validated a predictive model against real clinical trial data. The retrospective validation and prospective prediction of a Phase 2b study strengthen confidence, but the model has been tested on only two drugs.

When this study was published

Published in 2025, this represents current state-of-the-art in systems pharmacology for GLP-1-based peptide drug development, directly applicable to the many peptide agonists currently in clinical pipelines.

The bigger picture

With dozens of GLP-1 and multi-agonist peptide drugs in development pipelines, tools that can predict clinical outcomes from early data are extremely valuable. This model could help pharmaceutical companies quickly identify which peptide drug candidates are most likely to succeed, reducing the time and cost of bringing new diabetes medications to market. It represents the growing role of computational pharmacology in peptide drug development.

Questions still open

Can this model be extended to predict weight loss and cardiovascular outcomes, not just glucose control?
Will the model maintain accuracy for next-generation triple agonist peptides?
Could this modeling approach be used by regulators to support accelerated approval pathways for peptide drugs?

Common questions

How can a computer model predict drug effects from lab data?

The model takes two types of early information — how potently a peptide drug activates its receptor in lab tests (in vitro potency) and how the drug behaves in the body (pharmacokinetics) — and uses mathematical equations representing glucose metabolism to simulate what happens over weeks of treatment. By capturing the biology of insulin secretion, glucose uptake, and red blood cell turnover, it can forecast long-term HbA1c changes.

What is cotadutide and how does it differ from liraglutide?

Liraglutide activates only the GLP-1 receptor, while cotadutide is a dual agonist that activates both GLP-1 and glucagon receptors. Dual agonists may offer additional metabolic benefits beyond glucose control, including enhanced weight loss and improved liver fat metabolism. The fact that the model accurately predicted both drugs' effects demonstrates its versatility across different peptide agonist designs.

Read the original research

From In Vitro Efficacy to Long-Term HbA1c Response for GLP-1R/GlucagonR Agonism Using the 4GI-HbA1c Systems Model.

CPT: pharmacometrics & systems pharmacology, 14(9), 1515-1525

View on PubMed ↗Publisher’s paper ↗

Citation

Bosch, Rolien; Petrone, Marcella; Arends, Rosalin; Sijbrands, Eric J G; Hoefman, Sven; Snelder, Nelleke. (2025). From In Vitro Efficacy to Long-Term HbA1c Response for GLP-1R/GlucagonR Agonism Using the 4GI-HbA1c Systems Model.. CPT: pharmacometrics & systems pharmacology, 14(9), 1515-1525. https://doi.org/10.1002/psp4.70074