Controllable Protein Design by Prefix-Tuning Protein Language Models.

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Designing novel proteins with specific functions is crucial for tackling biomedical challenges.
• Existing methods for protein design can be limited in scope and efficiency.

Purpose of the Study:

• To introduce PrefixProt, a novel framework for controllable protein design.
• To enable multiobjective control over protein generation using adaptive virtual tokens.

Main Methods:

• PrefixProt utilizes prefix-tuning to learn virtual tokens as control tags.
• These virtual tokens are data-driven and combinatorially integrated for multiobjective control.
• The framework was validated through experiments on protein structure and function design.

Main Results:

• PrefixProt efficiently guides protein language models (ProtLM) by optimizing fewer parameters.
• It outperforms other parameter-efficient fine-tuning methods and text-guided ProtLMs, especially with limited data.
• The framework successfully generated proteins with multiple target properties, expanding design possibilities.

Conclusions:

• PrefixProt offers a robust and flexible framework for de novo protein design.
• The approach shows significant promise for applications in drug discovery and biomedicine.