Employing Machine Learning for the Prediction of Antimicrobial Resistance (AMR) Phenotypes

Antimicrobial resistance (AMR) poses a significant global public health challenge, responsible for the rise in hospital-acquired infections and increased levels of illness and death. The misuse and overuse of antibiotics have played a role in fostering drug resistance among pathogens, creating a pressing need for effective strategies to predict AMR phenotypes. Employing machine learning techniques has emerged as valuable tools in this endeavor, enabling the analysis of vast datasets to identify patterns and predict the resistance or susceptibility of microorganisms to specific antibiotics. The utilization of machine learning presents a promising approach to combat the growing threat of AMR, with the potential to significantly enhance patient outcomes. The objective of this study is to enhance AMR prediction employing machine learning techniques, leveraging insights from the cybersecurity domain due to the similarities between AMR and malware datasets. The approach involves employing k-mer frequency analysis and feature importance algorithms to extract significant features. The experimental outcomes highlight the following key findings: (1) Our approach demonstrates competitive performance, even with a small dataset; and (2) Utilizing 10-mers yields better outcomes than 7-mers. This research has shown that by applying cross -domain research methodologies and capitalizing on the shared characteristics among different datasets, the performance of AMR prediction can be improved.