GraphGPT: A Graph Enhanced Generative Pretrained Transformer for Conditioned Molecular Generation.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Artificial intelligence in drug discovery

Background:

• Condition-based molecular generation accelerates drug discovery by expanding virtual screening libraries.
• Existing methods may not fully capture complex molecular topological features or conditional generation requirements.

Purpose of the Study:

• To develop a novel computational model for condition-based molecular generation.
• To enhance the efficiency and accuracy of generating molecules with specific properties for drug discovery.

Main Methods:

• Combined molecular graph structures with sequential representations using a generative pretrained transformer (GPT) architecture.
• Incorporated graph structure information for better understanding of molecular topology.
• Utilized GPT's sequential contextual understanding for conditional molecular generation.

Main Results:

• The model efficiently generates molecules with desired properties, achieving valid and unique metrics close to 100%.
• Successfully preserved scaffold information during generation for scaffold-based tasks.
• Generated molecules exhibited low similarity while meeting specified property requirements.

Conclusions:

• The proposed GPT-based model offers an effective approach for condition-based molecular generation.
• This method significantly enhances the process of drug discovery by improving virtual screening and molecule design.
• The model's ability to maintain scaffold integrity and control molecular properties is crucial for targeted drug development.