MACI: A machine learning-based approach to identify drug classes of antibiotic resistance genes from metagenomic data

Novel methodologies are now essential for identification of antibiotic resistant pathogens in order to resist them. Here, we are presenting a model, MACI (Machine learning-based Antibiotic resistance gene-specific drug Class Identification) that can take metagenomic fragments as input and predict the drug class of antibiotic resistant genes. In our study, we trained a model using the Comprehensive Antibiotic Resistance Database, containing 5138 representative sequences across 134 drug classes. Among these classes, 23 dominated, contributing 85% of the sequence data. The model achieved an average precision of 0.8389 ± 0.0747 and recall of 0.8197 ± 0.0782 for these 23 drug classes. Additionally, it exhibited higher performance (precision and recall: 0.8817 ± 0.0540 and 0.8620 ± 0.0493) for predicting multidrug resistant classes compared to single drug resistant categories (0.7923 ± 0.0669 and 0.7737 ± 0.0794). The model also showed promising results when tested on an independent data. We then analysed these 23 drug classes to identify class-specific overlapping nucleotide patterns. Five significant drug classes, viz. "Carbapenem; cephalosporin; penam", "cephalosporin", "cephamycin", "cephalosporin; monobactam; penam; penem", and "fluoroquinolone" were identified, and their patterns aligned with the functional domains of antibiotic resistance genes. These class-specific patterns play a pivotal role in rapidly identifying drug classes with antibiotic resistance genes. Further analysis revealed that bacterial species containing these five drug classes are associated with well-known multidrug resistance properties.