Machine learning assisted identification of antibiotic-resistant Staphylococcus aureus strains using a paper-based...

Area of Science:

• Biomedical Engineering
• Microbiology
• Analytical Chemistry

Background:

• Staphylococcus aureus is a significant human pathogen, causing diverse infections and posing challenges due to antibiotic resistance and biofilm formation.
• Current methods for identifying antimicrobial-resistant (AMR) S. aureus strains like MRSA and VISA are often time-consuming and expensive.
• Previous work developed a paper-based sensor array with fluorescent dyes and machine learning for bacterial species and Gram status identification.

Purpose of the Study:

• To evaluate a paper-based ratiometric sensor array's ability to distinguish antibiotic-resistant Staphylococcus aureus strains and their biofilms.
• To assess the sensor array's performance in conjunction with machine learning algorithms for accurate and rapid identification of AMR S. aureus.

Main Methods:

• Utilized a paper-based sensor array with fluorescent sensor dyes (3-hydroxyflavone derivatives) pre-adsorbed on paper microzone plates.
• Employed machine learning algorithms including Linear Discriminant Analysis (LDA), neural networks, and support vector machines (SVM).
• Tested the sensor array on methicillin-resistant S. aureus (MRSA), methicillin-susceptible S. aureus (MSSA), and vancomycin-intermediate S. aureus (VISA) strains, including their biofilms.

Main Results:

• The sensor array differentiated MRSA from MSSA strains with 82.5% accuracy using LDA and neural networks.
• Support vector machines achieved 97.5% accuracy in classifying MRSA, MSSA, and VISA clinical isolates.
• The sensor array successfully discriminated AMR S. aureus biofilms with over 80% accuracy.

Conclusions:

• The paper-based sensor array demonstrates significant potential as a robust diagnostic tool for identifying drug-resistant S. aureus.
• The technology offers accurate, rapid, and easy identification of AMR S. aureus strains and biofilms in clinical settings.
• This approach could overcome limitations of current time-consuming and expensive diagnostic methodologies.