Identifying microbial protease allergens through protein language model-guided homology.

Area of Science:

• Microbiology
• Immunology
• Bioinformatics

Background:

• Emerging research implicates the gut, skin, and oral microbiomes in allergic diseases.
• Serine proteases (SPs) are increasingly recognized as potential allergens within these microbial communities.

Purpose of the Study:

• To develop and apply deep learning models to identify allergenic SPs in metagenomic data.
• To uncover novel candidate allergens in the human microbiome.

Main Methods:

• Utilized pre-trained protein language models (pLMs) to identify catalytic serine residues in serine hydrolases.
• Developed a deep learning framework to detect SP allergens across gene catalogs using conserved catalytic triads.
• Validated predicted allergens through experimental immunization in a model system.

Main Results:

• Successfully identified the catalytic serine residue, demonstrating pLMs' ability to capture structural information.
• Predicted a putative SP allergen similar to V8 protease and a cysteine protease resembling Der f 1.
• Experimental validation confirmed the allergenic potential of the identified proteases, inducing allergic responses.

Conclusions:

• The deep learning approach effectively identifies novel allergenic serine proteases in metagenomic data.
• This method expands the scope of allergen discovery beyond traditional techniques.
• Identified candidate allergens provide new targets for understanding and managing allergies.