An LC-MS-guided AI enabled serum molecule-interpretable SERS platform for exploratory metabolomic analysis in...

Label-free surface-enhanced Raman spectroscopy (SERS) offers a promising avenue for rapid metabolic phenotyping in complex biofluids, yet its translational potential is hindered by the challenge of deconvoluting overlapping spectral contributions and identifying specific metabolite signatures. Herein, we report a liquid chromatography-mass spectrometry (LC-MS)-guided AI enabled serum molecule-interpretable SERS Platform, termed AMI-SERS, which integrates SERS, untargeted metabolomics and machine learning (ML) for Alzheimer's disease (AD) patient serum exploratory metabolic profiling. ML has achieved excellent classification performance with an accuracy of 96.67% under leave-one-out cross-validation. A novel multi-source matching algorithm was devised to trace dominant spectral features directly to specific metabolite changes, uncovering hypoxanthine as a key pre-analytical confounder and nominating adenine as a potential AD biomarker. However, due to the limited sample size, these results are only applicable to exploratory analysis. In conclusion, the AMI-SERS platform establishes a molecular-resolvable, LC-MS-interpretable AI methodological framework for exploratory serum metabolomics and its clinical applicability requires large-scale independent validation.