An Interpretable, Thermodynamics-Based Deep Learning Framework for Predicting and Optimizing Drug Membrane...

Area of Science:

• Pharmacology and Computational Chemistry
• Drug Discovery and Development

Background:

• Cellular membranes act as selective barriers, making membrane permeability a critical factor in drug pharmacokinetics.
• Predicting and designing membrane permeability for novel therapeutics remains a significant challenge in drug development.

Purpose of the Study:

• To develop a novel thermodynamics-based deep learning framework for analyzing the structure-permeability relationship of small molecules.
• To create interpretable graph neural network models for predicting and designing drug membrane permeability.
• To demonstrate the framework's utility by designing a novel, highly permeable melatonin analog for nasal administration.

Main Methods:

• Utilized coarse-grained molecular dynamics simulations to determine membrane penetration thermodynamics for 8,239 compounds.
• Developed interpretable graph neural network (GNN) models to establish structure-permeability relationships.
• Designed and synthesized a novel melatonin analog (MT-A2) optimized for enhanced membrane permeability.

Main Results:

• The deep learning framework successfully predicted and analyzed the structure-permeability relationship.
• The designed melatonin analog, MT-A2, exhibited superior nasal absorption and prolonged brain retention compared to melatonin.
• MT-A2 demonstrated enhanced sleep efficacy in a proof-of-concept study.

Conclusions:

• The developed thermodynamics-based deep learning approach offers a promising strategy for predicting and designing membrane permeability.
• This method can significantly aid in the development of drugs with improved pharmacokinetic profiles.
• The successful design of MT-A2 highlights the potential of this framework in accelerating drug discovery for enhanced therapeutic outcomes.