Leveraging Graph Neural Networks for MIC Prediction in Antimicrobial Resistance Studies.

Antimicrobial resistance (AMR) poses a significant challenge in healthcare and public health, with organisms such as nontyphoidal Salmonella leading the way due to their escalating resistance to antimicrobial agents. This situation severely complicates the management and containment of diseases, highlighting the urgent need for more effective techniques to assess antimicrobial susceptibility. Conventional methods, including the broth microdilution technique for determining Minimum Inhibitory Concentrations (MICs), are time-consuming and require extensive manual effort. The advent of machine learning (ML) technologies offers a revolutionary approach to predicting MICs, thereby potentially increasing the efficacy of antimicrobial therapies. This paper explores the latest advancements in ML for MIC prediction, focusing on an innovative approach using Graph Neural Networks (GNNs), which could provide a novel insight into the correlation between gene fragment similarities and MIC values. Within this paper, we introduce the K-mer GNN, a novel GNN model designed for MIC prediction. The K-mer GNN model distinctively identifies and incorporates the similarities among k-mers, integrating these insights into GNN alongside k-mer features. This approach not only elevates the precision of MIC predictions but also sheds light on the genomic factors at the k-mer level that drive antimicrobial resistance.