Inverse design of 3d molecular structures with conditional generative neural networks

Area of Science:

• Computational chemistry
• Artificial intelligence in chemistry
• Molecular modeling

Background:

• Rational molecular design is a key challenge in chemistry.
• Generative neural networks (NNs) can generate novel molecules.
• Existing methods may struggle with sparse data or complex property targeting.

Purpose of the Study:

• To develop a conditional generative neural network for 3D molecular structures.
• To enable targeted sampling of molecules with specified chemical and structural properties.
• To address challenges in inverse design and property prediction.

Main Methods:

• Utilized a conditional generative neural network architecture.
• Developed a method agnostic to chemical bonding for flexible structure generation.
• Applied the model to domains with sparse reference calculations.

Main Results:

• Successfully generated 3D molecules with specified chemical motifs and composition.
• Discovered novel molecules exhibiting enhanced stability.
• Demonstrated joint targeting of multiple electronic properties, extending beyond the training data distribution.

Conclusions:

• The proposed conditional generative NN is effective for inverse molecular design.
• This approach facilitates the discovery of novel molecules with desired properties.
• The method shows promise for applications in materials science and drug discovery.