Evaluation of Marker Gene-Based In Silico Antimicrobial Resistance Prediction Tools

Area of Science:

• Microbiology
• Bioinformatics
• Public Health

Background:

• Antimicrobial resistance (AMR) surveillance is critical for clinical and public health.
• Traditional culture-based methods are standard, but culture-independent tools using 16S rRNA gene profiling are emerging.
• Evaluations of these predictive AMR tools are limited.

Purpose of the Study:

• To evaluate the accuracy of culture-independent functional prediction tools for antimicrobial resistance (AMR) detection.
• To compare the performance of PICRUSt2, Tax4Fun, and MicFunPred (16S rRNA-based) against shotgun sequencing (AMRFinderPlus).
• To assess the utility of 16S rRNA gene marker-based functional profiling for AMR surveillance.

Main Methods:

• 20 *E. coli* strains (10 carbapenem-resistant, 10 susceptible) were analyzed.
• AMR phenotype determined by Vitek2; DNA extracted for 16S rRNA (V3-V4) and shotgun sequencing.
• Bioinformatic pipelines: QIIM2 (16S rRNA), MetaPhlAn4 (shotgun); Functional prediction tools: PICRUSt2, Tax4Fun, MicFunPred (16S rRNA), AMRFinderPlus (shotgun).

Main Results:

• F1 scores for 16S rRNA predictive profilers were low: Tax4Fun (0.22), PICRUSt2 (0.12), MicFunPred (0.08).
• Tax4Fun demonstrated the highest F1 score among the 16S rRNA-based tools.
• Comparison of predicted AMR profiles against shotgun sequencing revealed generally low accuracy.

Conclusions:

• Current 16S rRNA gene-based functional prediction tools have limited accuracy for AMR detection.
• Specialized AMR databases and improved algorithms are necessary for reliable resistance prediction.
• Further research is needed to enhance the utility of culture-independent methods in AMR surveillance.