3DSMILES-GPT: 3D molecular pocket-based generation with token-only large language model

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery

Background:

• Generating valid and high-quality 3D molecules is crucial for drug discovery but faces challenges with existing computational methods.
• Current approaches often yield molecules with invalid configurations, poor drug-like properties, and slow generation times.

Purpose of the Study:

• To introduce 3DSMILES-GPT, a language-model-driven framework for efficient and accurate 3D molecular generation.
• To address limitations in molecular validity, conformation, drug-likeness, synthesizability, and generation speed.

Main Methods:

• Utilized a token-only language model approach for both 2D and 3D molecular representations.
• Pre-trained the model on millions of drug-like molecules, followed by fine-tuning with protein-pocket/molecule structural data.
• Employed reinforcement learning to optimize biophysical and chemical properties.

Main Results:

• 3DSMILES-GPT significantly outperforms existing methods in binding affinity, drug-likeness (QED), and synthetic accessibility score (SAS).
• Achieved a 33% enhancement in quantitative estimation of drug-likeness (QED) while maintaining state-of-the-art binding affinity.
• Demonstrated remarkable generation speed, averaging 0.45 seconds per molecule, a threefold increase over prior methods.

Conclusions:

• 3DSMILES-GPT offers a powerful and efficient solution for 3D molecular generation in drug discovery.
• The token-only, language-model-driven framework has the potential to accelerate the identification of novel drug candidates.