De novo functional protein sequence generation: overcoming data scarcity through regeneration and large language...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Protein Engineering

Background:

• Proteins are vital for life, with broad applications in medicine and materials.
• Protein design, particularly amino acid sequence design, is crucial for harnessing protein potential.
• Deep generative models show promise for protein sequence design but require substantial data.

Purpose of the Study:

• To develop a novel hierarchical model, ProteinRG, for generating functional protein sequences.
• To address the challenge of limited functional protein sequence data for training generative models.
• To enable efficient protein design using relatively small datasets.

Main Methods:

• ProteinRG employs a hierarchical approach, first generating a protein sequence representation.
• It leverages existing large protein sequence models for the initial representation.
• The model then generates a functional protein sequence based on this representation.

Main Results:

• Generated protein sequences demonstrated similarity to original sequences and functional consistency.
• Evaluations using multiple sequence alignment, t-SNE analysis, and 3D structure prediction supported the findings.
• ProteinRG exhibited superior performance compared to other generative models in protein sequence generation.

Conclusions:

• ProteinRG offers an effective solution for generating functional protein sequences from limited data.
• The model successfully maintains sequence similarity and functional integrity.
• ProteinRG represents a significant advancement in generative models for protein design.