# Multi-dimensional integration of gene expression, protein evidence, and serum autoantibodies for diagnostic modeling in esophageal squamous cell carcinoma

## TL;DR

Researchers developed a diagnostic model for early detection of esophageal squamous cell carcinoma by combining gene expression, protein evidence, and serum autoantibodies.

## Contribution

A novel multi-dimensional integrative approach combining transcriptomics, protein validation, and autoantibody profiling to build a diagnostic model for ESCC.

## Key findings

- Seven tumor-associated autoantibodies were significantly elevated in ESCC cases compared to controls.
- A support vector machine model achieved AUCs of 0.826 in training and 0.779 in independent testing.
- A web-based diagnostic tool was developed for clinical use.

## Abstract

Esophageal squamous cell carcinoma (ESCC) accounts for most esophageal cancer cases. This study implemented a multi-dimensional integrative approach to identify tumor-associated autoantibodies (TAAbs) and develop a diagnostic model for the early detection of ESCC.

The study comprised four phases: discovery, verification, modeling, and evaluation. Transcriptomic screening of public datasets and protein-level evidence from literature were integrated to identify candidate tumor-associated antigens (TAAs), followed by serological evaluation using enzyme-linked immunosorbent assay (ELISA) in 940 samples. Eight machine learning algorithms were assessed to develop the optimal diagnostic model.

In the discovery phase, transcriptomic analysis identified 26 differentially expressed genes in ESCC, of which ten genes encoding proteins with literature-supported evidence were selected as candidate TAAs for serological testing. Seven TAAbs were significantly elevated in ESCC cases compared with normal controls in the verification phase. In the modeling phase, six TAAbs (anti-CEP55, anti-CKS1B, anti-ECT2, anti-KIF2C, anti-SURV, and anti-TPX2) remained elevated in ESCC cases compared with both benign esophageal disease and normal controls. The support vector machine (SVM) model demonstrated the best diagnostic performance, achieving AUCs of 0.826 (95% CI: 0.776–0.876) in the training set and 0.741 (95% CI: 0.651–0.832) in the internal test set. In the evaluation phase, the SVM model was validated in an independent temporal test set (AUC 0.779, 95% CI 0.717–0.842). The web-based diagnostic tool is accessible at https://linzou.shinyapps.io/ESCC_SVM_Model/.

This multi-dimensional approach linking transcriptomic evidence, protein-level validation, and immunodiagnostic markers facilitated the development of a diagnostic model, which may hold promise for early detection of ESCC.