SERSome-based metabolic profiling of aqueous humor for cataract subtyping with molecule-level interpretability.

Cataract is among the leading cause of vision impairment worldwide and it is essential to uncover molecule-level characteristics of different cataract subtypes for personalized treatment. Compared to existing diagnostic techniques, surface-enhanced Raman spectroscopy (SERS) shows advantages in sensitive molecular fingerprinting, tractability and compactness, making it highly suitable for clinical adoption. To further address the current lack of SERS-based techniques for cataract subtyping, this work presented an integrated methodology for robust and interpretable metabolic profiling of aqueous humor. A two-step sample pretreatment protocol has been therein developed to efficiently extract metabolites while ensuring compatibility with subsequent SERS detection. SERSome strategy was employed to enable rapid and robust metabolic profiling. Molecule-level SERSome interpretation was achieved by matching its positive intra-correlation network with metabolites' SERS barcode. Multiple analytical methods including statistical comparison, interpretable machine learning, and association analysis have been leveraged, evidencing key metabolite biomarkers such as ergothioneine, hypoxanthine and uric acid associated with cataract subtyping and diabetic retinopathy. In a word, this study has not only established a systematic methodology for metabolic analysis of aqueous humor but also provided a usable tool for both fundamental research and clinical applications in different diseases with high rapidness, cost efficiency and scalability.