Abstract:
Antimicrobial resistance (AMR) presents an immediate threat to public health as microorganisms evolve to resist antimicrobial drugs, leading to challenging or untreatable infections. AMR-related costs could exceed 1 trillion dollars globally by 2050, surpassing major causes of death. In 2019, AMR directly caused 1.27 million deaths worldwide, surpassing mortality rates of HIV/AIDS and malaria. In particular, tuberculosis (TB) claimed 1.3 million lives in 2022, ranking as the second most prevalent infectious disease globally. This study introduces in silico approach utilizing deep learning to analyze the entire genome sequence of top pathogens such as Escherichia coli, Staphylococcus aureus, Klebsiella pneumonia, Streptococcus pneumonia, Pseudomonas aeruginosa, and Mycobacterium tuberculosis. The goal is to swiftly and accurately predict a pathogen’s resistance to specific drugs, eliminating the necessity for complex laboratory experiments. In addition to employing the standard label encoding technique, this study introduces three novel encoding methods for mutational data: transition- transversion encoding, codon frequency encoding, and gene based codon gain. These novel techniques are the major contribution of this study. Prediction of the AMR profile of the patient’s pathogen against various drugs is crucial before prescribing treatment. This approach helps in identifying the most appropriate drug that will be effective in treating the patient at the early stage, thereby reducing the likelihood of treatment failure caused by prescribing a drug to which the pathogen is resistant. While the model’s performance varies across different drugs and pathogens, it demonstrates the potential for application in antimicrobial resistance prediction.
