PARGT: a software tool for predicting antimicrobial resistance in bacteria.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Whole-genome sequencing is increasingly available, enabling new methods for antimicrobial resistance (AMR) gene identification.
• Traditional alignment-based methods face limitations, especially for unculturable pathogens.
• Previous work established a game-theory-based feature evaluation for AMR gene identification in Gram-negative bacteria.

Purpose of the Study:

• To extend the application of game-theory-identified features and machine learning to Gram-positive bacteria.
• To evaluate the accuracy of this approach for identifying specific antibiotic resistance genes.
• To provide a practical software tool for implementing the developed methodology.

Main Methods:

• Utilized a game-theory-based feature evaluation algorithm to identify protein characteristics.
• Coupled these game-theory-identified features with machine learning models.
• Applied the methodology to identify genes conferring resistance to bacitracin and vancomycin in Gram-positive bacteria.

Main Results:

• Achieved classification accuracies between 87% and 90% for bacitracin and vancomycin resistance genes.
• Demonstrated the effectiveness of the machine learning approach in Gram-positive bacteria.
• Developed a standalone software tool for the game-theory and machine learning model.

Conclusions:

• Machine learning coupled with game-theory-identified features is a powerful approach for detecting AMR genes in Gram-positive bacteria.
• The developed software tool offers a practical solution for identifying AMR genes, particularly for unculturable pathogens.
• This work contributes to combating antibiotic resistance through advanced computational methods.