Scalable and Generalizable Analog Design via Learning Medicinal Chemistry Intuition from Matched Molecular Pair...

Area of Science:

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

Background:

• Drug discovery relies on systematic chemical analog design, often guided by medicinal chemistry intuition.
• Existing models using matched molecular pairs (MMPs) are limited by data bias and lack generalizability.
• Representing medicinal chemistry intuition effectively for generative models remains a challenge.

Purpose of the Study:

• To develop a novel generative model approach for chemical analog design by focusing on matched molecular pair transformations (MMPTs) as the core unit.
• To create a generalizable and context-independent representation of medicinal chemistry intuition applicable across diverse drug discovery projects.
• To evaluate the performance of a transformation-centric foundation model (MMPT-FM) against traditional MMP-based models.

Main Methods:

• Developed a transformation-centric foundation model (MMPT-FM) using MMPTs as the fundamental unit of chemical modification.
• Trained and compared MMPT-FM with MMP-based generative models on a curated ChEMBL-derived dataset.
• Assessed model performance using real-world test cases from drug discovery patents, including within-patent and cross-patent evaluations.

Main Results:

• The MMPT-FM model demonstrated comparable or superior recall metrics across all evaluated test cases.
• The model showed particularly strong performance in generating low-frequency and previously unseen chemical transformations.
• The approach effectively captures and utilizes medicinal chemistry intuition in a scalable manner.

Conclusions:

• The transformation-centric approach offers a paradigm shift for learning and applying medicinal chemistry intuition in AI-driven drug discovery.
• MMPT-FM provides a scalable foundation model capable of encoding extensive medicinal chemistry knowledge for analog design.
• This method enhances the efficiency and effectiveness of designing novel chemical analogs, potentially accelerating drug discovery.