Scholarlink

<p>Cancer remains a major global health challenge, with early and accurate diagnosis crucial for improving outcomes. MicroRNAs (miRNAs), small non-coding RNAs involved in gene regulation, have shown promise as biomarkers due to their role in tumor progression and tissue specificity. However, identifying robust and interpretable miRNA signatures remains difficult. This study introduces NeoMiRX, a custom deep learning framework for cancer classification using miRNA expression profiles, offering both biological interpretability and downstream regulatory network integration. NeoMiRX was trained on miRNA expression data from TCGA for BRCA (breast), LUAD (lung adenocarcinoma), and UCEC (endometrial) cancers. The model is trained with a customised deep learning architecture which is suitable for high-dimensional biological data where traditional models may struggle, also provides interpretability to rank the top 15 miRNAs per cancer type and patients. These were mapped to validated miRNA-TF-gene regulatory networks. For BRCA, survival analysis via Kaplan-Meier and Cox regression was conducted. NeoMiRX achieved high accuracies: 99.38% (BRCA), 100% (LUAD), and 98.79% (UCEC), with ROC-AUCs of 0.9791, 1.0000, and 0.9946. SHAP revealed key oncogenic and tumor suppressor miRNAs, and network analysis identified hubs like TP53 and MYC. Several BRCA miRNAs correlated with survival. NeoMiRX enables accurate, interpretable miRNA-based cancer classification and biomarker discovery.</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

NeoMiRX an interpretable deep learning framework for miRNA-based cancer classification and regulatory network analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Ashimjyoti Nath,
  • Chandan Jyoti Kumar,
  • Raju Bharalee,
  • Sangita Baruah,
  • Keshab Nath,
  • Sk Mahmudul Hassan

摘要

Cancer remains a major global health challenge, with early and accurate diagnosis crucial for improving outcomes. MicroRNAs (miRNAs), small non-coding RNAs involved in gene regulation, have shown promise as biomarkers due to their role in tumor progression and tissue specificity. However, identifying robust and interpretable miRNA signatures remains difficult. This study introduces NeoMiRX, a custom deep learning framework for cancer classification using miRNA expression profiles, offering both biological interpretability and downstream regulatory network integration. NeoMiRX was trained on miRNA expression data from TCGA for BRCA (breast), LUAD (lung adenocarcinoma), and UCEC (endometrial) cancers. The model is trained with a customised deep learning architecture which is suitable for high-dimensional biological data where traditional models may struggle, also provides interpretability to rank the top 15 miRNAs per cancer type and patients. These were mapped to validated miRNA-TF-gene regulatory networks. For BRCA, survival analysis via Kaplan-Meier and Cox regression was conducted. NeoMiRX achieved high accuracies: 99.38% (BRCA), 100% (LUAD), and 98.79% (UCEC), with ROC-AUCs of 0.9791, 1.0000, and 0.9946. SHAP revealed key oncogenic and tumor suppressor miRNAs, and network analysis identified hubs like TP53 and MYC. Several BRCA miRNAs correlated with survival. NeoMiRX enables accurate, interpretable miRNA-based cancer classification and biomarker discovery.