MSDeepAMR: antimicrobial resistance prediction based on deep neural networks and transfer learning

Antimicrobial resistance (AMR) is a global health issue that requires early and effective treatments. Mass Spectrometry (MS) is widely used to identify bacterial species and detect AMR. However, deep learning analysis of AMR is still relatively new, with most models relying on manual preprocessing techniques. This study proposes a deep neural network, MSDeepAMR, to learn from raw mass spectra to predict AMR. The model was implemented for Escherichia coli, Klebsiella pneumoniae, and Staphylococcus aureus under different antibiotic resistance profiles. The results showed good classification performance, with an AUROC above 0.83 in most cases. This model can be used in laboratories without extensive sample collection capacity.