DepoScope: Accurate phage depolymerase annotation and domain delineation using large language models

Area of Science:

• Microbiology
• Bioinformatics
• Structural Biology

Background:

• Bacteriophages (phages) are viruses that infect bacteria and often utilize depolymerase enzymes to degrade bacterial polysaccharide structures.
• Accurate identification and domain annotation of depolymerases are hindered by their significant sequence diversity.
• Understanding depolymerase function is key to deciphering phage-host interactions.

Purpose of the Study:

• To develop a precise machine learning tool for identifying depolymerase sequences and their enzymatic domains.
• To enhance the annotation accuracy of phage-derived depolymerases.
• To improve the understanding of phage-host interactions mediated by depolymerases.

Main Methods:

• Developed DepoScope, a tool integrating a fine-tuned ESM-2 model with a convolutional neural network.
• Curated a specialized dataset from the INPHARED phage genome database.
• Created a dedicated polysaccharide-degrading domain database and applied sequential filters for dataset refinement.

Main Results:

• DepoScope achieves precise identification of depolymerase sequences.
• The tool accurately predicts the functional enzymatic domains within depolymerases.
• This represents the first approach combining sequence-level and amino-acid-level predictions for depolymerase analysis.

Conclusions:

• DepoScope offers a significant advancement in the accurate detection and functional domain identification of depolymerases.
• The tool can substantially improve our comprehension of phage-host interactions at the molecular level.
• DepoScope provides a valuable resource for researchers studying bacteriophages and their enzymatic machinery.