CCVAN Leverages Conditional Molecular Generation Through Conditional VAE and Wasserstein GAN

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Materials science

Background:

• Molecular generation is crucial for drug discovery and materials science.
• Existing methods face limitations in generating molecules with specific properties.
• Conditional generation of molecules with desired characteristics remains a challenge.

Purpose of the Study:

• To introduce CCVAN, a novel hybrid framework for conditional molecular generation.
• To integrate conditional variational autoencoder (CVAE) with Wasserstein generative adversarial network (WGAN).
• To demonstrate CCVAN's capability in generating molecules with specific properties for drug design.

Main Methods:

• Developed a hybrid architecture combining CVAE and WGAN, named CCVAN.
• Utilized CVAE for encoding molecular data into a latent space.
• Employed a shared decoder and WGAN's generator for molecular SMILES reconstruction.
• Trained the WGAN discriminator on both real and generated molecular SMILES for conditional generation.

Main Results:

• CCVAN successfully generates molecules with specific, desired properties.
• The framework demonstrates high validity and novelty in generated molecules compared to existing methods.
• CCVAN was applied to ligand-based and structure-based drug design, generating high-affinity molecules.
• Achieved accelerated compound discovery through flexible and effective molecular generation.

Conclusions:

• CCVAN is a flexible and effective framework for conditional molecular generation.
• The hybrid CVAE-WGAN approach enhances molecular generation for drug discovery and materials science.
• CCVAN accelerates the identification of high-binding-affinity molecules for targeted applications.
• The source code is publicly available, promoting further research and development.