CVAE

CVAE implements a conditional variational autoencoder (CVAE) for de novo molecular design that generates drug-like molecules conditioned on multiple molecular properties.

Key Features:

• Conditional generation: Imposes multiple molecular properties within a single generative process to produce molecules meeting specified attributes.
• Multi-property control: Provides fine-tuned control over individual molecular properties without compromising other specified properties.
• Out-of-distribution manipulation: Manipulates molecular properties beyond the range of existing datasets to explore novel chemical space.
• Latent space mapping: Maps multiple molecular properties onto a latent space within the CVAE framework.
• Proof-of-concept generation: Demonstrated generation of drug-like molecules satisfying five distinct target properties.
• De novo molecular design: Produces novel chemical structures intended for drug discovery and pharmaceutical research.

Scientific Applications:

• De novo drug discovery: Generation of candidate drug-like molecules with predefined target properties for pharmaceutical research.
• Design of complex therapeutic profiles: Concurrent specification of multiple desired characteristics to design compounds with complex profiles.
• Iterative molecular optimization: Enables iterative adjustment of specific molecular attributes while retaining other properties.
• Exploration of novel chemical space: Facilitates discovery of innovative compounds by extending property values beyond training data ranges.
• Targeted property generation: Produces molecules conditioned on explicitly specified molecular property targets.

Methodology:

Uses a conditional variational autoencoder framework that maps molecular properties onto a latent space and generates molecules conditioned on those properties, demonstrated by producing molecules meeting five target properties in a proof-of-concept study.