ASAP-ML: Antibiotic Susceptibility and Antibiogram Prediction With Machine Learning Methods

Area of Science:

• Genomic Medicine
• Computational Biology
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) is a growing global health crisis.
• Pathogen resistance mechanisms evolve rapidly, necessitating advanced detection methods.
• Genomic technologies now offer unprecedented capabilities for understanding AMR.

Purpose of the Study:

• To explore the use of machine learning with genomic data for predicting antibiotic resistance.
• To propose and evaluate a multi-model approach, ASAP (Antibiotic Susceptibility and Antibiogram Prediction), for generating antibiograms.
• To compare the predictive performance of ten distinct machine learning models.

Main Methods:

• Genomic data preprocessing using n-gram encoding for feature extraction.
• Implementation and evaluation of ten machine learning models: CNN, Nearest Neighbor, Random Forest, XGBoost, CatBoost, Naive Bayes, SVM, Light GBM, Gradient Boost, and Logistic Regression.
• Performance assessment using accuracy, recall, precision, and F1 scores.

Main Results:

• Machine learning models achieved high predictive accuracy for antibiotic resistance, reaching up to 0.99.
• Macro average recall exceeded 0.90 across models.
• XGBoost demonstrated the highest performance with 0.99 accuracy, while Naive Bayes showed 0.89 accuracy.

Conclusions:

• Machine learning combined with genomic sequences is a powerful tool for predicting AMR.
• The ASAP approach offers a promising alternative to manual antibiogram creation.
• This methodology can significantly aid healthcare professionals in making informed, less empirical antibiotic prescribing decisions.