Introduction

Age-related macular degeneration (AMD) is a leading cause of irreversible blindness worldwide, with significant challenges for early diagnosis and treatment.

Objectives

To identify new biomarkers that are important for the early diagnosis and monitoring of the severity/progression of AMD.

Methods

We investigated the diagnostic and monitoring potential of blood metabolites in a cohort of 547 individuals (167 healthy controls, 240 individuals with other eye diseases as eye disease controls, and 140 individuals with AMD) from 2 centers over three phases: discovery phase 1, discovery phase 2, and an external validation phase. The samples were analyzed via a mass spectrometry-based, widely targeted metabolomic workflow. In discovery phases 1 and 2, we built a machine learning algorithm to predict the probability of AMD. In the external validation phase, we further confirmed the performance of the biomarker panel identified by the algorithm. We subsequently evaluated the performance of the identified biomarker panel in monitoring the progression and severity of AMD.

Results

We developed a clinically specific three-metabolite panel (hypoxanthine, 2-furoylglycine, and 1-hexadecyl-2-azelaoyl-sn-glycero-3-phosphocholine) via five machine learning models. The random forest model effectively discriminated patients with AMD from patents in the other two groups and showed acceptable calibration (area under the curve (AUC) = 1.0; accuracy = 1.0) in both discovery phases 1 and 2. An independent validation phase confirmed the diagnostic model’s efficacy (AUC = 0.962; accuracy = 0.88). The three-biomarker panel model demonstrated an AUC of 1.0 in differentiating the severity of AMD via RF machine learning, which was consistent across both the discovery and external validation phases. Additionally, the biomarker concentrations remained stable under repeated freeze–thaw cycles (P > 0.05).

Conclusions

This study reveals distinct metabolite variations in the serum of AMD patients, paving the way for the development of the first routine laboratory test for AMD.

中文翻译：

---

用于年龄相关性黄斑变性早期诊断和监测的血清代谢物生物标志物

 介绍

年龄相关性黄斑变性 （AMD） 是全球不可逆失明的主要原因，对早期诊断和治疗提出了重大挑战。

 目标

确定对早期诊断和监测 AMD 严重程度/进展很重要的新生物标志物。

 方法

我们调查了来自 2 个中心的 547 名个体（167 名健康对照者，240 名患有其他眼病的个体作为眼病对照，以及 140 名患有 AMD 的个体）的血液代谢物的诊断和监测潜力，分为三个阶段：发现阶段 1、发现阶段 2 和外部验证阶段。通过基于质谱、广泛靶向的代谢组学工作流程对样品进行分析。在发现阶段 1 和 2 中，我们构建了一个机器学习算法来预测 AMD 的概率。在外部验证阶段，我们进一步确认了算法识别的生物标志物面板的性能。随后，我们评估了已确定的生物标志物面板在监测 AMD 进展和严重程度方面的性能。

 结果

我们通过五个机器学习模型开发了一个临床特异性的三代谢物组（次黄嘌呤、2-呋喃酰甘氨酸和 1-十六烷基-2-蝶草酰-sn-甘油-3-磷酸胆碱）。随机森林模型有效地区分了 AMD 患者与其他两组的专利，并在发现阶段 1 和 2 中都显示出可接受的校准 （曲线下面积 （AUC） = 1.0;准确度 = 1.0）。独立的验证阶段证实了诊断模型的有效性 （AUC = 0.962;准确性 = 0.88）。三生物标志物面板模型显示，通过射频机器学习区分 AMD 严重程度的 AUC 为 1.0，这在发现和外部验证阶段都是一致的。此外，生物标志物浓度在反复冻融循环下保持稳定 （P > 0.05）。

 结论

这项研究揭示了 AMD 患者血清中独特的代谢物变化，为开发 AMD 的首个常规实验室测试铺平了道路。