Scholarlink

Background <p>Uterine cancer (UC) is among the most fatal gynecological cancers, with increasing incidence and mortality and a five-year survival of about 90.35% in early stages versus 21.1% in advanced stages, highlighting the need for early detection. MicroRNAs (miRNAs) have emerged as promising noninvasive biomarkers for cancer diagnosis and prognosis, including uterine and endometrial cancers.</p> Objective <p>To identify differentially expressed miRNAs (demiRs) in uterine cancer and to construct a diagnostic miRNA panel using deep learning applied to The Cancer Genome Atlas (TCGA) dataset.</p> Methods <p>miRNA expression data comprising 20,513 miRNAs were extracted from the TCGA uterine cancer cohort. Differential expression analysis was performed to identify demiRs, yielding 347 candidates for further modelling. A deep neural network (DNN) was then trained on these 347 DemiRs to select the 20 most informative features. A generalized linear model (GLM) combined with receiver operating characteristic (ROC) curve analysis was applied to evaluate diagnostic performance and derive an optimal multi-miRNA panel.</p> Results <p>From the 20,513 profiled miRNAs, 347 DemiRs were identified as significantly dysregulated in uterine cancer compared with controls. DNN-based feature selection reduced this set to 20 top-performing miRNAs. Subsequent GLM and ROC analyses identified a two-miRNA panel, has-let-7c and has-mir-3622a, showing intense diagnostic discrimination for uterine cancer, despite the absence of significant associations between these miRNAs and available clinical variables.</p> Conclusion <p>This study demonstrates the utility of machine-learning approaches, particularly DNN-based feature selection combined with GLM and ROC analysis, for discovering candidate miRNA biomarkers in uterine cancer. The two-miRNA panel (has-let-7c and has-mir-3622a) shows potential as an early diagnostic tool, but requires extensive validation in independent cohorts and diverse laboratory settings before clinical implementation.</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

Combination of hsa.let.7c and hsa.mir.3622a as Potential miRNA in Uterine Cancer Diagnosis: RNA Seq and Deep Learning

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Tahmineh Aldaghi,
  • Mohammad Dashtiahangar,
  • Jan Muzik,
  • Diana Hosseinzadeh,
  • Ondrej Fanta,
  • Elham Nazari

摘要

Background

Uterine cancer (UC) is among the most fatal gynecological cancers, with increasing incidence and mortality and a five-year survival of about 90.35% in early stages versus 21.1% in advanced stages, highlighting the need for early detection. MicroRNAs (miRNAs) have emerged as promising noninvasive biomarkers for cancer diagnosis and prognosis, including uterine and endometrial cancers.

Objective

To identify differentially expressed miRNAs (demiRs) in uterine cancer and to construct a diagnostic miRNA panel using deep learning applied to The Cancer Genome Atlas (TCGA) dataset.

Methods

miRNA expression data comprising 20,513 miRNAs were extracted from the TCGA uterine cancer cohort. Differential expression analysis was performed to identify demiRs, yielding 347 candidates for further modelling. A deep neural network (DNN) was then trained on these 347 DemiRs to select the 20 most informative features. A generalized linear model (GLM) combined with receiver operating characteristic (ROC) curve analysis was applied to evaluate diagnostic performance and derive an optimal multi-miRNA panel.

Results

From the 20,513 profiled miRNAs, 347 DemiRs were identified as significantly dysregulated in uterine cancer compared with controls. DNN-based feature selection reduced this set to 20 top-performing miRNAs. Subsequent GLM and ROC analyses identified a two-miRNA panel, has-let-7c and has-mir-3622a, showing intense diagnostic discrimination for uterine cancer, despite the absence of significant associations between these miRNAs and available clinical variables.

Conclusion

This study demonstrates the utility of machine-learning approaches, particularly DNN-based feature selection combined with GLM and ROC analysis, for discovering candidate miRNA biomarkers in uterine cancer. The two-miRNA panel (has-let-7c and has-mir-3622a) shows potential as an early diagnostic tool, but requires extensive validation in independent cohorts and diverse laboratory settings before clinical implementation.