MolDecor: Leveraging Transformers to Decorate Bioactive Molecules for Property Optimization

Area of Science:

• Medicinal Chemistry
• Computational Drug Discovery
• Artificial Intelligence in Pharmacology

Background:

• Lead optimization is crucial for developing new therapeutics, requiring refinement of molecular structures.
• Improving pharmacological properties and drug-likeness is key for advancing lead molecules into clinical development.

Purpose of the Study:

• To develop an automated pipeline for lead optimization using a transformer-based model.
• To enhance specific molecular properties such as solubility and binding affinity.

Main Methods:

• A property-specific fragment (decorator) library was created.
• A BERT-based transformer model (MolDecor) learned fragment-scaffold relationships.
• Scaffolds were decorated with fragments to optimize lead molecule properties.

Main Results:

• MolDecor was trained on drug-like molecules and fine-tuned for solubility and affinity.
• The model generated novel, synthesizable molecules with improved properties.
• Successful optimization was demonstrated on anticancer, antimalarial, and kinase inhibitor molecules.

Conclusions:

• The developed pipeline offers an automated and unbiased approach to lead optimization.
• MolDecor effectively enhances critical drug properties, accelerating the drug discovery process.
• This method facilitates the generation of improved drug candidates with tailored characteristics.