Scholarlink

<p>This study presents a deep learning framework that integrates gene expression and proteomic-proxy features to predict breast cancer biomarkers ER, PR, and HER2. Using a dual-branch neural network, we process 3,273 samples from the GSE96058 dataset and apply SHAP-based explainability to uncover biologically meaningful feature contributions. The multi-omics model significantly outperforms a baseline CNN trained on imaging-derived features, achieving 91.4% accuracy, 0.931 AUC, and 0.894 F1-score across biomarkers, with a Kappa agreement of 0.87. These results demonstrate that combining transcriptomic and proteomic signals provides a more accurate and interpretable approach to biomarker prediction. Our findings underscore the potential of explainable multi-omics models in advancing personalized breast cancer diagnostics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating gene expression and proteomics for breast cancer biomarker prediction through a deep learning framework with SHAP-based explainability

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月27日

查看全文
  • Naim Ajlouni,
  • Abdelrahman Almassri

摘要

This study presents a deep learning framework that integrates gene expression and proteomic-proxy features to predict breast cancer biomarkers ER, PR, and HER2. Using a dual-branch neural network, we process 3,273 samples from the GSE96058 dataset and apply SHAP-based explainability to uncover biologically meaningful feature contributions. The multi-omics model significantly outperforms a baseline CNN trained on imaging-derived features, achieving 91.4% accuracy, 0.931 AUC, and 0.894 F1-score across biomarkers, with a Kappa agreement of 0.87. These results demonstrate that combining transcriptomic and proteomic signals provides a more accurate and interpretable approach to biomarker prediction. Our findings underscore the potential of explainable multi-omics models in advancing personalized breast cancer diagnostics.