Predicting antimicrobial resistance in Pseudomonas aeruginosa with machine learning-enabled molecular diagnostics

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Antibiotic resistance in bacteria limits therapeutic options.
• Genomic sequencing can predict antimicrobial resistance in some bacterial species.
• Optimizing diagnostics is crucial for managing drug-resistant infections.

Purpose of the Study:

• To develop predictive models for antimicrobial resistance in Pseudomonas aeruginosa using genomic and transcriptomic data.
• To identify biomarkers associated with resistance to four common antimicrobial drugs.
• To evaluate the diagnostic performance of machine learning classifiers.

Main Methods:

• Sequencing genomes and transcriptomes of 414 clinical Pseudomonas aeruginosa isolates.
• Training machine learning classifiers on gene presence/absence, sequence variation, and expression profiles.
• Assessing model sensitivity and predictive values for drug resistance.

Main Results:

• Machine learning models achieved high (0.8-0.9) to very high (>0.9) sensitivity and predictive values.
• Gene expression data improved diagnostic performance for most tested drugs, except ciprofloxacin.
• Identified genomic and transcriptomic biomarkers for antimicrobial resistance.

Conclusions:

• Developed accurate predictive models for antimicrobial resistance in Pseudomonas aeruginosa.
• Gene expression data enhances the prediction of antibiotic resistance.
• Results support the development of molecular susceptibility testing for improved clinical treatment.