Rectified Classifier Chains for Prediction of Antibiotic Resistance From Multi-Labelled Data With Missing Labels

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Genomic data analysis is crucial for predicting antimicrobial resistance (AMR), impacting healthcare.
• Machine learning (ML) shows promise for AMR prediction, but challenges remain for multi-drug resistance and incomplete datasets.
• Existing ML models often fail to identify biomarkers driving AMR predictions.

Purpose of the Study:

• To develop and evaluate a novel Rectified Classifier Chain (RCC) method for multi-drug AMR prediction.
• To address limitations in current methodologies, particularly with datasets containing missing labels.
• To enable the identification of biomarkers associated with AMR predictions.

Main Methods:

• Implementation of a Rectified Classifier Chain (RCC) model for multi-label classification of AMR.
• Utilizing annotated genomic sequence features as input for the RCC model.
• Comparison of the RCC model against other multi-label classification techniques, including a binary relevance model.

Main Results:

• The RCC method, using eXtreme Gradient Boosting (XGBoost) as a base model, significantly outperformed a comparable XGBoost-based binary relevance model.
• Achieved a 3.3% improvement in Hamming accuracy and a 7.8% improvement in F1-score compared to the next best model.
• Demonstrated the capability of the RCC method to identify informative biomarkers contributing to AMR prediction.

Conclusions:

• The proposed RCC method offers a robust solution for multi-drug AMR prediction from genomic data, effectively handling missing labels.
• This approach enhances diagnostic speed and treatment decisions in healthcare.
• The ability to identify biomarkers facilitates genome annotation and the discovery of novel AMR indicators.