Zero-shot segmentation using embeddings from a protein language model identifies functional regions in the human...

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Science

Background:

• Protein function is dictated by distinct units like folded domains and intrinsically disordered regions.
• Identifying and categorizing these segments from sequence is crucial for automatic protein annotation.
• Existing methods often rely on conserved sequence patterns.

Purpose of the Study:

• To present Zero-shot Protein Segmentation (ZPS), a novel method for identifying and categorizing protein segments using unsupervised protein language model embeddings.
• To demonstrate ZPS's ability to segment proteins without relying on conserved sequence patterns or requiring training.
• To evaluate ZPS's performance against established bioinformatics tools and its capacity for discovering novel functional regions.

Main Methods:

• Utilized embeddings from the unsupervised protein language model ProtT5.
• Developed Zero-shot Protein Segmentation (ZPS) to predict protein segment boundaries without parameter training or fine-tuning.
• Applied ZPS to the human proteome and analyzed segment embeddings for categorization.

Main Results:

• ZPS boundary predictions for the human proteome showed superior accuracy in reproducing UniProt annotations compared to established tools.
• ProtT5 embeddings of ZPS segments successfully categorized over 200 common UniProt annotations, including domains and intrinsically disordered regions.
• Introduced a novel visualization method for protein embeddings, aiding in the interpretation of distinct functional units.

Conclusions:

• ZPS offers a powerful, unbiased approach to identify and categorize both known and unannotated protein segments.
• The method successfully identified unannotated mitochondrion targeting signals and prion-like domains within intrinsically disordered regions.
• Analysis of protein segment embeddings holds significant potential for uncovering new insights into protein function, particularly for disordered and poorly understood regions.