目的：利用超高效液相色谱串联四极杆-静电场轨道阱高分辨质谱(ultra-high per formance liquid chromatography tandem quadrupole-electrostatic field orbitrap high resolution mass spectrometer, UHPLC- HRMS)代谢组学技术结合机器学习识别与糖尿病视网膜病变(diabetic retinopathy, DR)进展过程中的房水代谢差异，以寻找DR进展相关生物标志物。方法：本研究共纳入78例2型糖尿病(type 2 diabetes mellitus, T2DM)患者以及30名年龄性别匹配健康对照人群。使用UHPLC- HRMS检测所有患者及对照人群房水中的代谢物丰度，结合机器学习筛选T2DM和DR进展相关代谢物标志物并建立预测模型。结果：在校正混杂因素后，与健康人群对比1, 5-脱水山梨醇、硫酸十四烷基酯和N,N,N-三甲基-5-氨基戊酸在T2DM患者中表现出显著差异(均P < 0.05)；而N-乙酰色氨酸、亚油酰胺、油酰胺、棕榈酰胺、戊酸(游离脂肪酸(5:0)和琥珀酸与DR进展显著相关(均P < 0.001)。代谢通路分析表明，“缬氨酸、亮氨酸和异亮氨酸的生物合成”“精氨酸的生物合成”和“半胱氨酸及蛋氨酸代谢”是T2DM差异代谢途径。基于生物标志物的随机森林预测模型显示，差异代谢产物对T2DM和DR进展的预测准确率分别为81.3%和74%。结论：代谢组学结合机器学习方法有效揭示了T2DM及与DR进展相关的代谢特征，亚油酰胺和油酰胺有望成为DR进展的生物标志物，为DR的诊断和个体化治疗提供了新的可能性。

Objective: To identify aqueous humor metabolic profiles associated with the progression of type 2 diabetes mellitus (T2DM) and diabetic retinopathy (DR), aiming to discover potential biomarkers for DR progression. Ultra-high performance liquid chromatography tandem quadrupole-electrostatic field orbitrap high-resolution mass spectrometry (UHPLC-HRMS) will be utilized in conjunction with machine learning (ML) for comprehensive analysis. Methods: A total of 78 patients with T2DM and 30 age- and gender-matched healthy controls were included. UHPLC-HRMS was used to identify metabolites in the aqueous humor of all participants. ML was employed to screen for metabolites associated with T2DM and DR progression, and predictive models were established. Results: After adjusting for covariates, 1,5-anhydroglucitol, tetradecyl sulfate, and n,n,n-trimethyl-5-aminovaleric acid identified as significant indicators for T2DM compared to controls (all P < 0.05). N-acetyltryptophan, linoleamide, oleamide, palmitic amide, valeric acid(FFA(5:0), and succinic acid emerged as predictors for DR progression (all P < 0.001). Metabolic pathway analysis revealed that "valine, leucine and isoleucine biosynthesis", "arginine biosynthesis," and "cysteine and methionine metabolism" were the most enriched pathways for T2DM. Predictive models achieved R² values of 81.3%, and 74% for T2DM and DR progression, respectively. Conclusions: Metabolomic combined with ML effectively uncovered metabolic characteristics associated with T2DM and DR progression. Linoleamide and oleamide represent promising potential biomarkers for DR progression, offering new opportunities for diagnosis and personalized treatment of DR.