Predicting Peptide Permeability Across Diverse Barriers: A Systematic Investigation

Area of Science:

• Pharmacology and Drug Delivery
• Computational Chemistry
• Biotechnology

Background:

• Peptide therapeutics show great potential but face challenges with cell membrane permeability.
• Poor permeability limits intracellular delivery and oral drug development for peptides.

Purpose of the Study:

• To develop predictive models for peptide permeability using graph neural networks (GNNs) and machine learning.
• To systematically evaluate peptide permeability across diverse peptide types and cell lines.
• To identify key molecular features influencing peptide permeability.

Main Methods:

• Developed a novel GNN framework and machine learning algorithms for peptide permeability prediction.
• Evaluated models on natural, modified, linear, and cyclic peptides using Caco-2, RRCK, and PAMPA assays.
• Interpreted molecular structural features impacting permeability and the effects of cell lines and peptide modifications.

Main Results:

• Achieved high predictive accuracy (R² up to 0.708) for linear and cyclic peptides in Caco-2 and RRCK cell lines.
• GNN framework demonstrated superior performance with larger datasets, improving cyclic peptide prediction in PAMPA by ~0.32 R².
• Successfully identified critical molecular features and factors influencing peptide permeability.

Conclusions:

• The developed GNN framework offers a rapid and reliable strategy for assessing peptide permeability.
• These models aid in optimizing peptide drug delivery, preselecting peptides, and designing targeted peptide materials.
• The models are accessible via the user-friendly KNIME platform for broader research application.