Modeling Epoxidation of Drug-like Molecules with a Deep Machine Learning Network

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Metabolism

Background:

• Drug toxicity often arises from reactive metabolites, particularly epoxides, which bind to proteins.
• Epoxides are electrophilic cyclic ethers formed by cytochromes P450, often at aromatic or double bonds.
• Identifying the site of epoxidation (SOE) is crucial for predicting adverse events and designing safer drugs.

Purpose of the Study:

• To develop and validate a predictive model for identifying sites of epoxidation (SOE) in molecules.
• To utilize a deep convolution network based on existing metabolism and reactivity algorithms.
• To assess the model's performance in predicting SOEs and distinguishing them from other molecular sites.

Main Methods:

• A deep convolution network was employed, building upon an algorithm for predicting cytochrome P450 metabolism (XenoSite).
• The model was trained on a database of 702 epoxidation reactions.
• Predictions were made at both atom and molecule levels to identify SOEs.

Main Results:

• The model achieved 94.9% AUC in identifying SOEs and 79.3% AUC in distinguishing epoxidized from non-epoxidized molecules.
• It effectively differentiated aromatic/double bond SOEs from other aromatic/double bonds (92.5% and 95.1% AUC).
• The model also distinguished SOEs from sites of sp(2) hydroxylation with 83.2% AUC.

Conclusions:

• This novel epoxidation prediction model accurately identifies sites of epoxidation.
• The model shows significant potential for application in the development of safer drugs by predicting and preventing toxic metabolite formation.
• The developed epoxidation model is publicly available for research use.