Interpretable Ensemble Learning Predicts Antibiotic Resistance in Treponema denticola Using Expert Classifiers.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology and Machine Learning
• Infectious Diseases and Microbiology

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat, increasing healthcare costs and mortality.
• Periodontal infections, often involving pathogens like Treponema denticola, are a growing concern for AMR.
• Targeted therapies are crucial for managing AMR in periodontal disease.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting and classifying AMR genomic sequences in Treponema denticola.
• To identify the most effective ML model for accurate AMR classification in this key periodontal pathogen.
• To explore the potential of ML in informing targeted therapeutic strategies against periodontal infections.

Main Methods:

• Utilized UniProt FASTA sequences for T. denticola to investigate AMR.
• Employed BioPython library for data retrieval and preprocessing within a Jupyter Notebook environment.
• Compared four ML classification models (Random Forest, SVM, Gradient Boosting, Neural Network) and a Voting Classifier, optimizing hyperparameters and using fivefold cross-validation.

Main Results:

• The Voting Classifier demonstrated superior performance, achieving the highest test accuracy (96.46%) and F1-score (0.9646).
• Support Vector Machine (SVM) and Neural Network models also showed high accuracy (95.58%).
• The Voting Classifier exhibited robustness with a low log loss of 0.1504, indicating a good balance between accuracy and model calibration.

Conclusions:

• The Voting Classifier is highly effective for classifying AMR genomic sequences in T. denticola.
• Interpretable ML approaches show promise for advancing AMR research in periodontal pathogens.
• Accurate AMR prediction can enhance clinical decision-making, optimize antibiotic selection, and support public health surveillance.