Generative Artificial Intelligence Optimization of Albumin Binders: Coumarin and Fatty Acid Derivatives

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Artificial Intelligence

Background:

• Previous dual combinations for human serum albumin (HSA) binding, based on 4-hydroxycoumarin and dodecanedioic acid, faced optimization challenges limiting chemical diversity and affinity.
• Empirical selection and extensive screening methods are often suboptimal for improving synergistic binding agents.

Purpose of the Study:

• To establish a systematic artificial intelligence (AI) framework for optimizing synergistic HSA binding combinations.
• To improve molecular affinity and expand chemical diversity while preserving core chemotypes using AI-driven methods.

Main Methods:

• Trained a machine learning classifier on HSA binding data to serve as a scoring function.
• Utilized reinforcement learning-driven scaffold decoration with LibINVENT for goal-directed generation of coumarin and fatty acid derivatives.
• Synthesized and validated candidate molecules using surface plasmon resonance, molecular docking, and molecular dynamics simulations.

Main Results:

• Optimized molecules exhibited nanomolar HSA binding affinity, surpassing original ligands.
• AI framework successfully generated diverse coumarin and fatty acid derivatives with enhanced binding.
• Mechanistic studies revealed additional stabilizing interactions and favorable energetics contributing to improved affinity.
• A warfarin-derived coumarin analogue showed no detectable anticoagulant activity.

Conclusions:

• The AI-guided framework provides a practical route for expanding chemical diversity and enhancing affinity in synergistic HSA binding combinations.
• This approach demonstrates the potential of integrating computational optimization with experimental validation for efficient drug discovery.
• The developed method can be applied to optimize other synergistic drug combinations.