Integrating Generative Pretrained Transformer and Genetic Algorithms for Efficient and Diverse Molecular Generation

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Molecular generation models are vital for accelerating drug development.
• Current models offer either high performance with poor interpretability (deep learning) or better interpretability with limited performance (heuristic algorithms).

Purpose of the Study:

• To introduce an innovative molecular generation model, CCMol, integrating Generative Pretrained Transformer (GPT) and Genetic Algorithms (GA).
• To achieve effective and innovative molecular structures through a hybrid approach.
• To enhance early-stage drug discovery by expanding chemical validity.

Main Methods:

• Utilized structure-based drug design incorporating ligand and protein primary structures.
• Integrated GPT for initial molecular generation.
• Employed GA for iterative optimization of physicochemical and biological properties.
• Validated the model by generating molecules targeting GLP1, WRN, and JAK2 proteins.

Main Results:

• CCMol demonstrated comparable performance to advanced models across multiple indicators.
• Outperformed baseline models in structure diversity and drug-related properties.
• Successfully generated novel candidate drug molecules targeting key disease proteins.

Conclusions:

• CCMol offers a powerful and interpretable approach to molecular generation in computer-aided drug design.
• The model shows significant potential for developing novel and effective drug candidates.
• CCMol is particularly suitable for enhancing chemical diversity in early-stage drug discovery.