Designing optimized drug candidates with Generative Adversarial Network.

Area of Science:

• Computational Chemistry
• Artificial Intelligence in Drug Discovery
• Molecular Modeling

Background:

• Traditional drug design faces challenges like low efficacy, high cost, and time consumption.
• Deep Learning (DL) offers de novo drug design capabilities but often neglects crucial stereochemistry.
• Generating stereochemically accurate molecules is vital for targeted drug development.

Purpose of the Study:

• To propose a DL framework for de novo drug design that explicitly incorporates stereochemistry.
• To generate novel, realistic molecules with specific desired properties and high binding affinity.
• To address limitations of existing DL models in drug design regarding stereochemical precision.

Main Methods:

• A Feedback Generative Adversarial Network (GAN) framework integrating Encoder-Decoder, GAN, and Predictor models.
• Utilizing an Encoder-Decoder for molecular representation conversion into latent space vectors.
• Incorporating a feedback loop for property-based molecule evaluation and a non-dominated sorting genetic algorithm for optimization.

Main Results:

• The framework successfully generated realistic, novel molecules spanning the chemical space.
• The Encoder-Decoder model achieved 99% reconstruction accuracy, including stereochemical data.
• Generated molecules demonstrated high binding affinity to target receptors (Kappa Opioid, Adenosine A2A) and high diversity (0.88 internal, 0.94 external).

Conclusions:

• The proposed DL framework effectively generates stereochemically accurate and novel drug-like molecules.
• The approach enhances de novo drug design by optimizing for multiple objectives and chemical space exploration.
• This method shows significant potential for accelerating pharmaceutical research and development.