Automated and modular protein binder design with BinderFlow.

Area of Science:

• Computational biology
• Protein engineering
• Artificial intelligence in drug discovery

Background:

• Deep learning models have advanced de novo protein design, enabling the creation of novel proteins with specific functions, such as artificial protein binders.
• Current protein design workflows are computationally intensive and require specialized infrastructure and expertise, limiting accessibility.

Purpose of the Study:

• To present BinderFlow, an open, structured, and parallelized pipeline for automating end-to-end protein binder design.
• To enhance the accessibility, scalability, and reproducibility of generative protein design for both exploratory and production-level research.

Main Methods:

• Developed BinderFlow, a batch-based pipeline with a modular design for automated protein binder generation.
• Integrated BFmonitor, a web-based dashboard for real-time monitoring, evaluation, and selection of protein binder candidates.
• Demonstrated BinderFlow's utility through automated design campaigns generating diverse, high-confidence candidates.

Main Results:

• BinderFlow automates the entire protein binder design process, reducing computational demands and user intervention.
• The pipeline's architecture supports live monitoring and efficient resource utilization alongside other GPU-intensive tasks.
• Automated campaigns successfully generated numerous high-confidence protein binder candidates ready for experimental validation.

Conclusions:

• BinderFlow and BFmonitor significantly improve the accessibility, scalability, and reproducibility of generative protein design.
• The integrated system streamlines research by simplifying complex computational workflows for protein binder discovery.
• The open-source availability of BinderFlow facilitates broader adoption and further development in the field.