ProRefiner: an entropy-based refining strategy for inverse protein folding with global graph attention

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Inverse Protein Folding (IPF) is crucial for protein design, aiming to create sequences for a target structure.
• Current IPF methods often struggle with noisy residue predictions in local contexts.
• This noise can limit the accuracy and effectiveness of designed protein sequences.

Purpose of the Study:

• To develop an improved method for Inverse Protein Folding that overcomes limitations of existing approaches.
• To enhance the accuracy and efficiency of designing protein sequences compatible with given backbone structures.
• To introduce a novel model that leverages denoised context for superior sequence generation.

Main Methods:

• An entropy-based residue selection method was developed to filter noisy input residue contexts.
• ProRefiner, a memory-efficient global graph attention model, was introduced to process denoised contexts.
• The model's performance was evaluated on multiple sequence design benchmarks across various settings.

Main Results:

• The proposed entropy-based selection method effectively removes noise from residue contexts.
• ProRefiner achieved state-of-the-art performance on established protein sequence design benchmarks.
• The model demonstrated success in redesigning Transposon-associated transposase B, with promising variants for gene editing.

Conclusions:

• The developed method significantly improves the quality of input context for protein sequence design.
• ProRefiner offers a powerful and efficient tool for state-of-the-art protein sequence design.
• The findings highlight the potential of ProRefiner in practical applications like enhancing gene editing tools.