Hybrid quantum-classical machine learning for generative chemistry and drug design.

Area of Science:

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

Background:

• Deep generative models accelerate drug discovery but face challenges with vast molecular structural spaces.
• Hybrid quantum-classical architectures offer a potential solution to overcome these limitations.

Purpose of the Study:

• To develop a compact hybrid quantum-classical deep generative model for drug discovery.
• To demonstrate the feasibility of using current quantum annealers for generating novel drug-like molecules.

Main Methods:

• Implemented a compact discrete variational autoencoder (DVAE) with a reduced-size Restricted Boltzmann Machine (RBM) in its latent layer.
• Trained the DVAE model on a subset of the ChEMBL database of biologically active compounds.
• Utilized a D-Wave quantum annealer for model implementation and training.

Main Results:

• Generated 2331 novel chemical structures.
• The generated molecules exhibited medicinal chemistry and synthetic accessibility properties comparable to those in the ChEMBL database.
• The model's size was optimized to fit on existing quantum computing hardware.

Conclusions:

• The study validates the use of hybrid quantum-classical deep generative models for drug discovery.
• Existing quantum computing devices can serve as testbeds for future drug discovery applications.
• This approach shows promise for accelerating the identification of new therapeutic compounds.