Scholarlink

Background <p>Detection of ovarian cancer at early stages is well known to impact the survival rate significantly, but the sensitivity of current diagnostic techniques is low. Thus, there is a need to develop new diagnostic techniques. This study aimed to analyze an existing gene expression dataset and identify potential molecular biomarkers using bioinformatics, statistical, and machine learning techniques.</p> Results <p>Gene expression datasets were retrieved from the ArrayExpress database and analyzed to identify differentially expressed genes using the limma package in R. Feature selection using the Boruta algorithm yielded 14 potential biomarker genes, including 8 upregulated (<i>HOXB2, FOLR1, NEK2, KRT18, WFDC2, EHF, KLK6, CELSR1</i>) and 6 downregulated (<i>SGK1, CLSTN2, STAR, OSR2, FOXL2, ADAMTS5</i>) candidates. Functional annotation revealed that many of these dysregulated genes are involved in biological processes such as cell–cell adhesion and response to mineralocorticoid corticosterone mechanisms known to influence cancer progression through modulation of inflammation, cellular stress, and the tumor microenvironment. A machine learning-based predictive model developed using these 14 genes achieved an accuracy of 89.5% on the test dataset. <i>CELSR1</i>, <i>SGK1</i>, and <i>STAR</i> emerged as novel molecular biomarkers among the identified genes. Progression-free survival analysis using the Kaplan–Meier Plotter indicated a significant association between these biomarkers and poor prognosis, suggesting their potential role in disease progression.</p> Conclusions <p>The gene set identified in this study shows promise as diagnostic biomarkers for ovarian cancer. In particular, <i>CELSR1</i>, <i>SGK1</i>, and <i>STAR</i> may serve as novel candidates for future therapeutic targeting. The machine learning model developed for ovarian cancer prediction is publicly available at <a href="https://github.com/PGlab-NIPER/OC_Pred.git">https://github.com/PGlab-NIPER/OC_Pred.git</a>, offering a potential tool for advancing diagnostic strategies in oncology.</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

Integrative bioinformatics, statistical, and survival analysis reveals potential molecular biomarkers of ovarian cancer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月11日

查看全文
  • Pradnya Kamble,
  • Hardeep Sandhu,
  • Veena Puri,
  • Anju Sharma,
  • Prabha Garg

摘要

Background

Detection of ovarian cancer at early stages is well known to impact the survival rate significantly, but the sensitivity of current diagnostic techniques is low. Thus, there is a need to develop new diagnostic techniques. This study aimed to analyze an existing gene expression dataset and identify potential molecular biomarkers using bioinformatics, statistical, and machine learning techniques.

Results

Gene expression datasets were retrieved from the ArrayExpress database and analyzed to identify differentially expressed genes using the limma package in R. Feature selection using the Boruta algorithm yielded 14 potential biomarker genes, including 8 upregulated (HOXB2, FOLR1, NEK2, KRT18, WFDC2, EHF, KLK6, CELSR1) and 6 downregulated (SGK1, CLSTN2, STAR, OSR2, FOXL2, ADAMTS5) candidates. Functional annotation revealed that many of these dysregulated genes are involved in biological processes such as cell–cell adhesion and response to mineralocorticoid corticosterone mechanisms known to influence cancer progression through modulation of inflammation, cellular stress, and the tumor microenvironment. A machine learning-based predictive model developed using these 14 genes achieved an accuracy of 89.5% on the test dataset. CELSR1, SGK1, and STAR emerged as novel molecular biomarkers among the identified genes. Progression-free survival analysis using the Kaplan–Meier Plotter indicated a significant association between these biomarkers and poor prognosis, suggesting their potential role in disease progression.

Conclusions

The gene set identified in this study shows promise as diagnostic biomarkers for ovarian cancer. In particular, CELSR1, SGK1, and STAR may serve as novel candidates for future therapeutic targeting. The machine learning model developed for ovarian cancer prediction is publicly available at https://github.com/PGlab-NIPER/OC_Pred.git, offering a potential tool for advancing diagnostic strategies in oncology.