| dc.description.abstract | Methicillin-Resistant Staphylococcus aureus (MRSA) was first identified in the UK in 1961. It is a type of bacteria resistant to multiple antibiotics and has a very high mortality and morbidity rate, earning it the label of a “superbug.” Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) provides a faster alternative for microbial identification. When combined with machine learning, it significantly enhances the accuracy and efficiency of microbial identification. MRSA strains from different geographic regions exhibit specific sequence types distribution. Previous studies have primarily focused on predicting MRSA resistance within a single region, neglecting the impact of geographic differences on resistance patterns, leading to decreased model performance due to differences in the sources of MRSA datasets. To address this issue, our study utilized over 25,000 data samples from Taiwan and Switzerland. We employed deep learning methods combined with federated learning and transfer learning to train the model across regions, improving its adaptability and accuracy. The results showed that the AUROC and AUPRC in the independent test sets reached over 0.82 and 0.7, respectively. Through shapley value and differential expression analysis, we identified several key features associated with MRSA specific peaks, such as the 2415 peak in the 2410-2419 range representing PSM-mec and the 6593 peak in the 6590-6599 range representing the SA1452 protein, and conducted a comparative analysis of resistance and susceptible samples from Taiwan and Switzerland. Finally, from a macroscopic perspective, we explored the global distribution and prevalence of MRSA sequence types, finding correlations with our results, such as ST59-V and ST239-III in Taiwan and ST22-IV in Switzerland or Europe. These findings not only enrich our understanding of the global distribution of MRSA but also emphasize the importance of using localized data to improve the accuracy of resistance predictions. | en_US |