Genome-Wide Mutation Scoring for Machine-Learning-Based Antimicrobial Resistance Prediction.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Genomic prediction of antimicrobial resistance (AMR) aids patient outcomes.
• Machine learning models using k-mer counts are common for AMR prediction but face challenges with high-dimensional data.
• Limited training data hinders model performance and interpretability.

Purpose of the Study:

• To enhance machine learning models for AMR prediction by incorporating biologically relevant features.
• To evaluate the utility of predicted protein mutation functional impact as a novel feature for AMR prediction.

Main Methods:

• Genomic sequences from 19,521 isolates across nine pathogens were analyzed.
• The PROVEAN tool was used to predict the functional impact of mutations.
• These impact predictions were integrated as new features into machine learning models.

Main Results:

• The addition of functional impact features significantly improved AMR prediction accuracy for *Pseudomonas aeruginosa*, *Citrobacter freundii*, and *Escherichia coli*.
• Balanced accuracy improved by an average of 6.0% for these three pathogens.
• The approach demonstrated enhanced predictive performance over standard k-mer based models.

Conclusions:

• Feature engineering with predicted mutation functional impact offers a promising strategy to improve genomic AMR prediction.
• This method addresses limitations of high-dimensional k-mer data and limited training sets.
• The findings support the translation of genomic insights into improved clinical diagnostics for antimicrobial resistance.