ProtAlign-ARG: antibiotic resistance gene characterization integrating protein language models and alignment-based...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Antibiotic resistance gene (ARG) evolution and spread is a global health crisis.
• Current DNA sequencing methods for ARG detection have limitations in identifying novel variants.
• Protein language models offer potential but require extensive training databases.

Purpose of the Study:

• To introduce ProtAlign-ARG, a novel hybrid model for improved ARG detection and classification.
• To leverage protein language models and alignment scoring for enhanced ARG identification capacity.
• To expand ARG detection capabilities beyond existing alignment-based and standalone protein language model approaches.

Main Methods:

• Developed ProtAlign-ARG, a hybrid model integrating a pre-trained protein language model with an alignment scoring method.
• Utilized raw protein language model embeddings for accurate ARG classification.
• Incorporated alignment-based scoring (bit scores, e-values) for classification when model confidence is low.
• Extended ProtAlign-ARG for predicting ARG functionality and mobility.

Main Results:

• ProtAlign-ARG demonstrated high accuracy in identifying and classifying ARGs.
• The model showed superior recall compared to existing ARG identification tools.
• ProtAlign-ARG effectively predicted ARG functionality and mobility, showcasing its versatility.
• Comprehensive comparisons confirmed ProtAlign-ARG's superior performance over standalone models.

Conclusions:

• ProtAlign-ARG represents a significant advancement in ARG detection and classification from genomic data.
• The hybrid approach overcomes limitations of existing methods, particularly for novel ARG variants.
• This model enhances surveillance and understanding of antibiotic resistance spread.