Learning Relationships Between Chemical and Physical Stability for Peptide Drug Development

Area of Science:

• Pharmaceutical Development
• Computational Chemistry
• Drug Stability

Background:

• Chemical and physical stabilities are critical in pharmaceutical development.
• Chemical stability is assessed by potency and degradation products.
• Physical stability is often measured using Thioflavin-T.

Purpose of the Study:

• To develop a machine learning model for predicting chemical stability over time.
• To reduce the time and resources required for drug product stability studies.
• To explore the relationship between formulation conditions, time, and chemical stability.

Main Methods:

• Developed a multilayer perceptron (MLP) for total degradation prediction.
• Developed a random forest (RF) model for potency prediction.
• Evaluated model performance on a random test set using R² and MAE.

Main Results:

• MLP achieved R² of 0.945 and MAE of 0.421 for total degradation.
• RF achieved R² of 0.908 and MAE of 1.435 for potency.
• Including physical stability data reduced MAE to 0.148 for degradation and 0.705 for potency.

Conclusions:

• Chemical stability can be effectively modeled using machine learning techniques.
• A significant relationship exists between the physical and chemical stability of peptides.
• Machine learning offers a promising approach to accelerate stability assessments in drug development.