Transcriptome-Based Machine Learning Models to Predict Antimicrobial Resistance in Pseudomonas aeruginosa

Area of Science:

• Microbiology
• Genomics
• Computational Biology

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat, necessitating advanced detection methods.
• Pseudomonas aeruginosa is a key pathogen known for its adaptability and contribution to multidrug resistance.

Purpose of the Study:

• To assess the efficacy of transcriptome-based machine learning (ML) models in predicting AMR in P. aeruginosa.
• To identify gene expression signatures associated with antibiotic resistance phenotypes.

Main Methods:

• Transcriptomic profiles from 414 clinical isolates of P. aeruginosa were analyzed.
• ML algorithms were employed to predict resistance to ceftazidime, ciprofloxacin, meropenem, and tobramycin.
• Distinct and shared transcriptomic signatures were identified for each antibiotic.

Main Results:

• ML models demonstrated high predictive accuracy for AMR, with the tobramycin model achieving 98.8% accuracy and 100% sensitivity.
• Specific pathways like biofilm formation, membrane transport, and virulence were enriched in resistant isolates.
• A core set of 10 genes and 10 mRNAs were consistently deregulated across all tested antibiotics, indicating shared resistance mechanisms.

Conclusions:

• Transcriptome-based signatures show promise for developing mechanism-guided diagnostic assays for AMR in P. aeruginosa.
• Transcriptome-based ML models are effective tools for predicting antimicrobial resistance.
• Further validation in independent cohorts is recommended.