Cyclic peptide membrane permeability prediction using deep learning model based on molecular attention transformer.

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• Membrane permeability is a key challenge in developing cyclic peptide drugs.
• Experimental permeability testing is expensive and time-consuming.
• Accurate in silico prediction tools for cyclic peptide membrane permeability are limited.

Purpose of the Study:

• To develop an advanced computational model for predicting cyclic peptide membrane permeability.
• To improve the efficiency and accuracy of drug development pipelines.

Main Methods:

• Development of the Cyclic Peptide Membrane Permeability (CPMP) model using the Molecular Attention Transformer (MAT) framework.
• Validation of the model's predictive performance using PAMPA, Caco-2, RRCK, and MDCK permeability assays.
• Ablation studies to assess the contribution of individual MAT architecture components.
• Analysis of the impact of data pre-training and conformational optimization on model accuracy.

Main Results:

• The CPMP model achieved high determination coefficients (R ² ) for various permeability predictions: 0.67 (PAMPA), 0.75 (Caco-2), 0.62 (RRCK), and 0.73 (MDCK).
• The model's performance surpassed traditional machine learning and graph-based neural network approaches.
• Ablation experiments confirmed the effectiveness of the MAT architecture components.

Conclusions:

• The CPMP model offers a robust and accurate in silico solution for predicting cyclic peptide membrane permeability.
• This tool can significantly accelerate the drug discovery and development process by reducing reliance on experimental methods.