Protein A-like Peptide Design Based on Diffusion and ESM2 Models.

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Protein design is crucial for biotechnology, with historical reliance on structure-centric methods like Rosetta3.
• The advent of AlphaFold2 shifted focus towards deep learning for protein design.
• Generating novel, functional proteins remains a significant challenge.

Purpose of the Study:

• To develop a new method for generating functional proteins using diffusion models and the ESM2 protein language model.
• To demonstrate controlled generation of new protein sequences from minimal input data.
• To validate the biological activity of generated sequences.

Main Methods:

• Utilized a diffusion model, commonly used in image and language generation, for protein sequence generation.
• Employed the ESM2 protein language model, trained on extensive sequence data, for enhanced biological relevance.
• Applied the combined models to the Protein A-like peptide as a model system.
• Experimentally verified biological activity using methods like BLI affinity testing.

Main Results:

• Successfully generated novel peptide sequences using the diffusion and ESM2 models.
• Demonstrated the capability to create sequences from minimal input data.
• Experimentally confirmed the biological activity of the designed peptides, including affinity.

Conclusions:

• Developed a novel, effective method for protein design integrating diffusion models and protein language models.
• The proposed strategy offers a new approach to address challenges in generic protein generation.
• This work advances the field of computational protein design with a focus on functional outcomes.