Transformer-Based Molecular Generative Model for Antiviral Drug Design.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Artificial intelligence in medicine

Background:

• Simplified Molecular Input Line Entry System (SMILES) lacks human readability and editability for molecular design.
• International Union of Pure and Applied Chemistry (IUPAC) nomenclature offers superior human readability and editability, facilitating analogue-based drug design.
• Current methods for antiviral drug design, particularly analogue-based approaches, can be improved by leveraging IUPAC's functional group level editing capabilities.

Purpose of the Study:

• To introduce TransAntivirus, a data-driven, self-supervised pretraining generative model for designing antiviral candidate analogues.
• To enable select-and-replace edits and property-driven molecular conversions for enhanced drug design.
• To demonstrate the model's superiority over control methods in generating novel, valid, unique, and diverse molecular structures.

Main Methods:

• Development of a novel data-driven self-supervised pretraining generative model named TransAntivirus.
• Utilizing IUPAC nomenclature for human-oriented molecular editing and generation of new molecules.
• Employing chemical space analysis and property prediction for design and optimization of nucleoside and non-nucleoside analogues.
• Conducting case studies for nucleoside and non-nucleoside analogue design against coronavirus disease (COVID-19).

Main Results:

• TransAntivirus significantly outperformed control models in novelty, validity, uniqueness, and diversity of generated molecules.
• The model demonstrated excellent performance in the design and optimization of both nucleoside and non-nucleoside antiviral analogues.
• Four candidate lead compounds against COVID-19 were successfully screened through case studies.

Conclusions:

• TransAntivirus is a powerful framework for accelerating antiviral drug discovery and development.
• The model's ability to perform knowledge-based molecular design from the functional group level is crucial for analogue development.
• The successful identification of COVID-19 candidate lead compounds validates the practical applicability of TransAntivirus in real-world drug design scenarios.