ARGai 2.0: A Feature Engineering Enabled Deep Network Model for Antibiotic Resistance Gene and Strain Identification in E. coli

Abstract

Escherichia coli (E. coli) is a group of bacteria that cause infections in the gastrointestinal (GI) tract and urinary tract (UTIs). The rise in antimicrobial resistance (AMR) due to antibiotic-resistant genes (ARGs) linked to E. coli strains that cause UTIs poses a significant threat. Identifying ARGs and resistant strains is crucial for effective treatment. To identify ARGs and classify resistant strains in E. coli utilizing gene expression (GE) data with advanced computational techniques such as feature engineering and transfer learning (TL). In TL, knowledge acquired by the baseline model is transferred to the target domain (BI-LSTM-GRU). ARGai 2.0 utilizes the Synthetic Minority Over-sampling Technique (SMOTE) for over-sampling the GE dataset to evaluate the effectiveness of the proposed TL framework. Our proposed ARGai 2.0 model achieved a higher classification accuracy of 14% compared to 1D CNN and 11% compared to BI-LSTM-GRU individually. The analysis revealed that genes associated with the nitrate reductase operon (narU, narV, narW, narY, and narZ) exhibit high connectivity and interaction scores, indicating their central role in nitrate metabolism. This aligns with the high enrichment FDR of 3.07E-10 and fold enrichment of 229.28 for pathways related to nitrate reductase complex and nitrite transmembrane transporter activity. ARGai 2.0 successfully detected the significant genes responsible for antibiotic resistance and classified the resistant strains. The gene network analysis highlights the central role of nitrate metabolism genes, while peripheral genes like ansP and yncG are involved in more specialized functions.