Membrane Permeating Macrocycles: Design Guidelines from Machine Learning.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Large molecules beyond the Rule of Five (bRo5) are vital drug candidates but often fail to cross cell membranes.
• Poor cell permeability limits access to intracellular targets and oral bioavailability for bRo5 compounds.

Purpose of the Study:

• To develop a machine learning model for predicting passive membrane permeation rates in bRo5 macrocyclic compounds.
• To guide the design of novel, cell-permeable macrocycles for medicinal chemistry and chemical biology.

Main Methods:

• A random forest (RF) machine learning model was developed using over 1000 bRo5 macrocyclic compounds.
• The model utilizes easily calculable chemical features including polar surface area, lipophilicity, and descriptors of molecular flexibility and "chameleonicity".

Main Results:

• The RF model significantly outperformed a multiple linear regression model and previous studies using the same data.
• Key predictive features identified include polar surface area, octanol-water partitioning coefficient, hydrogen-bond donors, and topological distances between heteroatoms.

Conclusions:

• The developed RF model accurately predicts passive membrane permeation for bRo5 macrocycles.
• The model provides design guidelines to enhance membrane permeability, facilitating the development of new therapeutics.