Abstract
Purpose :
Glaucoma is a leading cause of permanent blindness worldwide. Identifying biomarkers is crucial for the early detection and monitoring of disease progression. Herein, we utilized machine learning algorithms to develop and validate potential metabolite biomarkers for diagnosing glaucoma from metabolomic data.
Methods :
A total of seven glaucoma individuals and nine age-matched healthy controls were recruited. For enhanced metabolite analysis, we utilized the Thermo Scientific Orbitrap IQ-X Mass Spectrometer on frozen serum samples and employed two chromatographic separation modes (Amide column and T3 column). We utilized three tree-based machine learning models on metabolomic data to identify metabolic alterations and potential glaucoma biomarkers.
Results :
A total of 3192 metabolite features were detected using Amide-positive(Amide-Pos), Amide-negative(Amide-Neg), T3-positive(T3-Pos), and T3-negative(T3-Neg) analytical modes. Among them, a total of 651 compounds were identified with higher confidence subject to biological interpretation. The Random Forest model achieved the highest accuracy of 0.87 among the three models and exhibited superior performance (p-value = 0.019). After the Shapley Additive exPlanations (SHAP) analysis, N, N'-Dicyclohexylurea was found to be the most significant feature for the classification of glaucoma patients.
Conclusions :
We successfully utilized machine learning and advanced spectrometry to identify potential biomarkers for glaucoma diagnosis. Notably, N, N'-Dicyclohexylurea emerged as a significant feature in distinguishing glaucoma patients from healthy controls. Scientific studies have established a connection between N, N'-Dicyclohexylurea and both ocular surface disease and systemic blood pressure. However, further investigation is required to fully understand the potential relationship between N, N'-Dicyclohexylurea, and glaucoma. Although further studies with larger sample sizes are needed, this approach allowed us to find a potential metabolite biomarker that could predict the early stage of glaucoma.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.