癌症的早期检测在临床实践中仍然是一个未满足的需求，迫切需要高诊断灵敏度和特异性的生物标志物。在这里，我们试图鉴定由超级增强子 (SE) 驱动的基因编码的分泌蛋白作为食管鳞状细胞癌 (ESCC) 的诊断生物标志物。我们对多个数据集进行了综合分析，包括 ChIP-seq 数据、分泌组数据、CCLE数据和 GEO 数据来筛选 SE 驱动基因编码的分泌蛋白。使用 ELISA，我们通过少量临床样本进一步鉴定了上调的分泌蛋白，并在多中心验证阶段进行了验证（测试队列中 345 个，验证队列中 231 个）。接受者操作特征曲线用于计算诊断准确性。应用名为梯度增强机（GBM）的人工智能（AI）方法构建模型，以提高诊断准确性。检测到血清EFNA1和MMP13，与正常对照相比，ESCC患者的血清EFNA1和MMP13水平显着升高。通过结合 EFNA1、MMP13、癌胚抗原、Cyfra21-1 和鳞状细胞癌抗原建立的综合五生物标志物面板 (iFBPanel) 在测试和验证队列中对于 ESCC 的 AUC 分别为 0.881 和 0.880。重要的是，iFBPanel 在检测早期 ESCC 患者方面也表现出良好的性能（0.872 和 0.864）。此外，iFBPanel在AI技术的进一步赋能下，在早期ESCC中表现出了出色的诊断性能（0.927和0.907）。我们的研究表明血清EFNA1和MMP13可能有助于ESCC检测，并提供了易于使用的检测模型这可能有助于早期 ESCC 的诊断。版权所有 © 2023 作者。由 Elsevier B.V. 出版。保留所有权利。

Early detection of cancer remains an unmet need in clinical practice, and high diagnostic sensitivity and specificity biomarkers are urgently required. Here, we attempted to identify secreted proteins encoded by super-enhancer (SE)-driven genes as diagnostic biomarkers for esophageal squamous cell carcinoma (ESCC).We conducted an integrative analysis of multiple data sets including ChIP-seq data, secretome data, CCLE data and GEO data to screen secreted proteins encoded by SE-driven genes. Using ELISA, we further identified up-regulated secreted proteins through a small size of clinical samples and verified in a multi-centre validation stage (345 in test cohort and 231 in validation cohort). Receiver operating characteristic curves were used to calculate diagnostic accuracy. Artificial intelligence (AI) method named gradient boosting machine (GBM) were applied for model construction to enhance diagnostic accuracy.Serum EFNA1 and MMP13 were identified, and showed significantly higher levels in ESCC patients compared to normal controls. An integrated Five-Biomarker Panel (iFBPanel) established by combining EFNA1, MMP13, carcino-embryonic antigen, Cyfra21-1 and squmaous cell carcinoma antigen had AUCs of 0.881 and 0.880 for ESCC in test and validation cohorts, respectively. Importantly, the iFBPanel also exhibited good performance in detecting early-stage ESCC patients (0.872 and 0.864). Furthermore, the iFBPanel was further empowered by AI technology which showed excellent diagnostic performance in early-stage ESCC (0.927 and 0.907).Our study suggested that serum EFNA1 and MMP13 could potentially assist ESCC detection, and provided an easy-to-use detection model that might help the diagnosis of early-stage ESCC.Copyright © 2023 The Authors. Published by Elsevier B.V. All rights reserved.