MolGAN-QRL: a hybrid framework for molecule generation using quantum-enhanced reinforcement learning.

Area of Science:

• Computational Chemistry
• Artificial Intelligence in Drug Discovery
• Quantum Computing Applications

Background:

• Drug discovery faces challenges in identifying novel candidates.
• Generative AI models like GANs show promise but suffer from limited chemical coverage and mode collapse.
• Existing methods struggle to optimize molecular validity, uniqueness, and drug-likeness simultaneously.

Purpose of the Study:

• To develop MolGAN-QRL, a hybrid quantum-classical framework to improve de novo molecular generation.
• To address limitations of classical Generative Adversarial Networks (GANs) in drug discovery.
• To enhance chemical coverage, validity, and uniqueness of generated molecules.

Main Methods:

• Developed MolGAN-QRL, integrating a variational quantum circuit (VQC) into the reinforcement learning reward module of MolGAN.
• Implemented a hybrid reward mechanism for optimizing molecular properties.
• Utilized quantum-enhanced reinforcement learning for guided exploration in molecular generation.

Main Results:

• MolGAN-QRL demonstrated superior generative performance compared to classical MolGAN.
• Achieved up to a 16-fold increase in unique and valid generated compounds.
• Showcased effective mitigation of mode collapse through enhanced uniqueness and validity.

Conclusions:

• Hybrid quantum-classical methods, specifically MolGAN-QRL, offer significant advancements in generative chemistry.
• Quantum-enhanced reward modeling is valuable for overcoming mode collapse in molecular generation.
• MolGAN-QRL shows potential for accelerating drug discovery by generating higher quality novel molecules.