MolT5-Linker: A Transformer-Based Sequence- and Putative Site-Guided Generative Framework for Antibody-Drug Conjugate...

Area of Science:

• Computational chemistry
• Biotechnology
• Drug discovery

Background:

• Generating effective antibody-drug conjugate (ADC) linkers is complex, requiring careful balancing of stability, payload release, and component compatibility.
• Existing methods face challenges in achieving optimal linker properties for diverse ADC applications.

Purpose of the Study:

• To introduce MolT5-Linker, a Transformer-based generative framework for rational ADC linker design.
• To improve the chemical compatibility and medicinal chemistry properties of generated ADC linkers.

Main Methods:

• Developed a Transformer-based generative framework (MolT5-Linker) conditioned on antibody sequences and payload molecular structures.
• Integrated a Graph Attention Network (GAT)-based Attachment Site Attention Network to infer attachment sites.
• Fine-tuned the model on a custom ADC dataset to optimize linker generation.

Main Results:

• MolT5-Linker achieved a generation validity of 0.8986, with good molecular recovery (0.5802) and uniqueness (0.5438).
• Generated linkers demonstrated medicinal chemistry property distributions (e.g., molecular weight, lipophilicity, TPSA) consistent with established ADC linkers.
• The framework successfully improved compatibility between generated linkers and ADC components.

Conclusions:

• MolT5-Linker provides a robust computational framework for the rational generation of ADC linker candidates.
• The model's ability to balance linker properties and ensure compatibility advances ADC design strategies.
• This approach holds promise for accelerating the development of novel antibody-drug conjugates.