What is it about?
Type 2 diabetes mellitus (T2DM) is a prevalent chronic metabolic disorder characterized by progressive systemic changes, where early screening and rapid staging of obesity-induced T2DM stages (obesity [OBs], impaired glucose tolerance [IGTs], diabetes mellitus [DMs]) remains challenging due to the lack of stage-specific bio-markers. Herein, we developed a robust non-targeted SERS-AI platform that integrated surface enhanced Raman scattering (SERS) and interpretable machine learning (ML) algorithms for multi-stage screening and staging of obesity induced T2DM. We analyzed 72,600 Raman fingerprint signals derived from plasma samples including 60 OBs, 60 IGTs, 62 DMs and 60 healthy controls using multiple candidate ML models. Among them, Quadratic SVM (QSVM) achieved the highest quadrupleclassification prediction accuracy of 94.5% after kernel-parameter optimization. For the post-hoc interpretability, SHapley Additive exPlanations (SHAP) analysis identified the top 15 high-weight Raman shifts associated with stage-specific biochemical constituents, manifesting many possible T2DM-related bio-structures. Comparative evaluation also showed QSVM outperform neural networks in stability and interpretability under non-big-data conditions. As a result, the different stages of T2DM progression can be staged clearly with biochemical interpretability and insight, showing great potential of our platform.
Featured Image
Photo by Viktor Forgacs on Unsplash
Why is it important?
Our work has five Highlights: 1. The first research conducting multi-stage analysis of obesity-induced T2DM (as a developmental disorder) via SERS-AI platform with a dynamic perspective. 2. Traversing majority common ML models with biochemical SHAP interpretation. 3. Using 242 clinical samples for dataset construction and method verification. 4. Fabricating composite AgNWs substrates for non-targeted SERS acquisition with great generalizability. 5. Comparing the conditioned performance of SVM and neural networks comprehensively.
Perspectives
It's a meaningful work and we hope it can provide good insight and reference. We are confident that our research establishes a clinically translatable SERS-AI platform capable of distinguishing obesity-induced T2DM stages with low costing and great biochemical interpretability, representing a significant advance in precision diabetology.
Qixin Miao
Fudan University
Read the Original
This page is a summary of: Multi-stage screening and staging of obesity-induced T2DM via non-targeted SERS-AI platform with biochemical interpretability, Sensors and Actuators B Chemical, December 2025, Elsevier,
DOI: 10.1016/j.snb.2025.138373.
You can read the full text:
Contributors
The following have contributed to this page
