Predicting bacterial resistance from whole-genome sequences using k-mers and stability selection

Area of Science:

• Genomics
• Computational Biology
• Microbiology

Background:

• Predicting bacterial antibiotic resistance from whole-genome sequences is feasible.
• Current methods often rely on detecting known resistance genes or mutations.
• An alternative approach using statistical learning without prior genetic information is needed.

Purpose of the Study:

• To develop a supervised statistical learning method for predicting bacterial antibiotic resistance phenotypes.
• To utilize k-mer based genotyping and logistic regression without prior knowledge of resistance factors.
• To identify a minimal yet predictive set of k-mers for resistance prediction.

Main Methods:

• Employed a k-mer based genotyping scheme.
• Utilized a logistic regression model to combine k-mers into a probabilistic model.
• Applied stability selection with Lasso penalty and resampling to identify predictive k-mers.

Main Results:

• Achieved predictive performance equivalent to state-of-the-art methods on public datasets for two bacterial species.
• Developed extremely sparse models involving only 1 to 8 k-mers per antibiotic.
• Demonstrated that the models are fast and easy to evaluate on new genomes.

Conclusions:

• Stability selection is a powerful approach for investigating bacterial genotype-phenotype relationships.
• The developed method offers an efficient and accurate way to predict antibiotic resistance.
• This proof of concept opens new avenues for genomic epidemiology and antimicrobial stewardship.