Shape-Based Generative Modeling for de Novo Drug Design.

Area of Science:

• Computational chemistry
• Machine learning
• Drug discovery

Background:

• De novo molecular design is crucial for identifying novel drug candidates.
• Existing methods often lack efficient ways to incorporate 3D structural information and pharmacophoric features.
• Exploring novel chemical space with lead-like properties remains a significant challenge.

Purpose of the Study:

• To develop a novel machine learning pipeline for de novo molecular generation.
• To guide the design process using a seed compound's 3D shape and pharmacophoric features.
• To generate novel molecules with lead-like properties in unexplored chemical regions.

Main Methods:

• A machine learning approach inspired by generative models in image analysis.
• Utilizing a variational autoencoder to perturb the 3D representation of a seed compound.
• Employing convolutional and recurrent neural networks to generate SMILES (Simplified Molecular Input Line Entry System) token sequences.

Main Results:

• Successful generation of novel molecular scaffolds and functional groups.
• Demonstration of a de novo design pipeline guided by shape-based features.
• Exploration of previously uncharted areas of chemical space while maintaining lead-like characteristics.

Conclusions:

• The proposed machine learning approach enables the generative design of novel, lead-like molecules.
• Shape-based features can effectively guide the de novo design process.
• This method expands the accessible chemical space for drug discovery.