Artificial Intelligence Based Approaches for Prediction of Antimicrobial Resistance in Ruminant Host Pathogens

Abstract A growing global health concern is antimicrobial resistance (AMR), especially in livestock such as ruminants, where overuse of antibiotics leads to resistant infections. Predicting AMR in these hosts accurately is crucial for enhancing treatment plans and reducing hazards to the public's health. Machine Learning (ML) and Deep Learning (DL) based framework for predicting AMR in ruminant-associated bacterial pathogens is presented in this study. The sequences of 190 strains of Staphylococcus aureus, Enterococcus faecalis, and Escherichia coli were collected from different international bioprojects based on how the strains responded to five widely used antibiotics—Methicillin, Penicillin, Ampicillin, Ciprofloxacin, and Gentamicin. Features like GC content, k-mer frequencies, and open reading frame (ORF) statistics were extracted from genomic sequence data from susceptible and resistant strains. These characteristics were used to train and assess a number of deep learning (CNN, RNN) and machine learning (RF, SVM, XGBoost) models. Furthermore, a transformer-based architecture called DNABERT was fine-tuned on DNA sequences using 6-mer tokenization in order to capture contextual nucleotide patterns. The two models XGBoost and DNABERT were found to be the best with the precision 84.8% and 86.7% respectively. RF, CNN and RNN models also showed competitive performance. Further, a web-based application have been developed to make it easier for users to upload genomic sequences and receive real-time resistance/susceptibility forecasts. The findings show that feature-based and sequence-based models have the ability to accurately predict resistance in animal microbiomes by combining ML, DL, and transformer architectures, providing a substitute for reference dependent and database-dependent best-hit based methods.