LDI-MS with Co3O4@Au nanohybrid matrix enables machine learning diagnosis of esophageal cancer
The study reports a novel hollow Co3O4@Au nanocomposite matrix for laser desorption/ionization mass spectrometry (LDI-MS) combined with machine learning to differentiate esophageal cancer from benign esophagitis. Investigators acquired serum metabolic profiles from 278 participants, including healthy controls and patients with esophagitis or cancer. A Random Forest model achieved an AUC of 0.989 for distinguishing cancer from esophagitis, and a refined 10-feature m/z panel sustained an AUC of 0.978. The platform offers low background interference and high throughput, highlighting clinical mass spectrometry as a viable noninvasive approach for esophageal malignancy screening and differential diagnosis.
The original study
Noble Metal-Metal Oxide Nanohybrids as a High-Performance LDI-MS Matrix for Machine Learning-Driven Metabolic Diagnosis of Esophageal Cancer.
- Authors
- Wang C, Shi M, Peng P, Qu Y, Yu X, Bu Y, et al.
- Journal
- Analytical chemistry
- Type
- Journal Article
- PMID
- 42461696
Original abstract
Esophageal cancer represents a global health challenge with a notably high incidence and poor prognosis, necessitating the development of rapid, noninvasive diagnostic methodologies. In this study, we present a high-throughput metabolomics platform leveraging a hollow-structured Co3O4@Au nanocomposite as a matrix for laser desorption/ionization mass spectrometry (LDI-MS) to diagnose esophageal cancer and differentiate it from benign esophagitis. Synthesized via a metal-organic framework (MOF) derivation strategy followed by the in situ reduction of gold nanoparticles, the Co3O4@Au matrix exhibits strong photoelectric properties, high charge separation efficiency, and robust tolerance to complex biological environments. This enables the direct, rapid extraction of serum metabolic profiles with low background interference. Leveraging this platform, serum metabolic profiles were acquired from a clinical cohort of 278 participants, including 116 healthy controls, 80 patients with esophagitis, and 82 patients with esophageal cancer. Integrated machine learning algorithms, notably the Random Forest model, demonstrated robust diagnostic performance, achieving a high AUC value for distinguishing diseased individuals from healthy controls and an AUC of 0.989 for differentiating esophageal cancer from esophagitis. Furthermore, a streamlined diagnostic panel comprising 10 core m/z features was selected, which sustained a high predictive accuracy (AUC = 0.978) and was biologically validated via SHAP interpretability analysis. This work establishes a high-throughput machine learning-driven strategy, offering a potential noninvasive tool for the mass screening and precision differential diagnosis of esophageal malignancies.