Machine learning for identifying tumor stemness genes and developing prognostic model in gastric cancer

Guoxing Li, Yunpeng Chen, Youyang Hu, Wenjing Zhao, Yun-Yan Lu, Fujian Wan, Zhijun Wu, Xiangqian Wang, Qiying YuAging 2024 · DOI

TL;DR

This study develops a machine learning-based prognostic model (TSRGPM) for gastric cancer by identifying key tumor stemness genes (TSKGs). The model effectively stratifies patient risk and identifies CDC25A as a crucial driver of malignant phenotypes, offering potential for personalized treatment strategies.

What Problem Does This Paper Address?

Gastric cancer is a challenging disease with a poor prognosis, often linked to tumor stem cells (CSCs) that drive treatment resistance and recurrence. Identifying genes associated with tumor stemness is crucial for developing better prognostic tools and targeted therapies. However, the heterogeneity of gastric cancer and the complex role of CSCs make this a difficult task. This paper aims to address this by systematically identifying TSKGs and building a robust prognostic model to improve patient outcomes.

Methods

The researchers employed a multi-pronged approach:

  • Tumor Stemness Gene Identification: They combined Weighted Gene Co-expression Network Analysis (WGCNA) with Gene Set Enrichment Analysis (ssGSEA) to identify potential tumor stemness genes.
  • Machine Learning for TSKG Selection: Five distinct machine learning algorithms (SVM, KNN, Decision Tree, Logistic Regression, and Random Forest) were used to screen the genes identified in the previous step and select the most important Tumor Stemness Key Genes (TSKGs).
  • Prognostic Model Development: The selected TSKGs were used to construct the Tumor Stemness Risk Genes Prognostic Model (TSRGPM). The model's efficacy was evaluated using various statistical methods.
  • Validation and Investigation: The model's prognostic value was validated using Cox regression and ROC curve analysis. Immune infiltration patterns were analyzed using CIBERSORT. Finally, CDC25A, identified as a key TSKG, was further investigated through bioinformatics and in vitro experiments to understand its role in cell proliferation, cell cycle, and tumor stemness.

Datasets

  • TCGA (The Cancer Genome Atlas): Used for initial screening of differentially expressed genes and for developing and testing the prognostic model.
  • GEO (Gene Expression Omnibus) dataset (GSE84437): Used as an independent test dataset to validate the TSRGPM.

Key Insights and Ideas

  • Machine Learning for Precision Oncology: The study highlights the power of integrating multiple machine learning algorithms to identify critical genes and build predictive models in complex diseases like gastric cancer.
  • CDC25A as a Key Driver: The identification and validation of CDC25A as a significant contributor to gastric cancer's malignant phenotype, cell proliferation, and DNA replication is a major finding. This provides a specific target for further research and potential therapeutic intervention.
  • TSRS as an Independent Prognostic Indicator: The developed Tumor Stemness Risk Score (TSRS) effectively differentiates patient risk and is an independent predictor of overall survival, offering a valuable tool for clinical decision-making.
  • Immune Microenvironment Link: The correlation between TSRS and immune cell infiltration, particularly monocytes, suggests a complex interplay between tumor stemness and the tumor microenvironment that influences prognosis.

Key Results

  • The study successfully identified a set of Tumor Stemness Key Genes (TSKGs) using a combination of WGCNA and ssGSEA, followed by machine learning algorithms.
  • The developed Tumor Stemness Risk Genes Prognostic Model (TSRGPM) demonstrated significant prognostic value, with higher Tumor Stemness Risk Scores (TSRS) correlating with poorer overall survival in both TCGA and GEO datasets.
  • TSRS was found to be an independent prognostic indicator and showed a significant correlation with lymph node metastasis.
  • Immune infiltration analysis revealed associations between TSRS and monocytes.
  • In vitro experiments confirmed that inhibiting CDC25A expression reduced gastric cancer cell viability, altered cell cycle progression, and decreased CD44 expression, indicating its role in proliferation and stemness.

Limitations and Future Work

  • Validation of Patient Distribution: The study acknowledges limitations in validating the distribution of patients within the high TSRS group across different N classifications due to constraints in obtaining clinical samples.
  • Further Molecular and Animal Studies: The authors emphasize the need for comprehensive molecular and animal experiments to fully elucidate the intricate mechanisms governing the role of TSKGs in tumor stemness regulation. This will guide future research and the development of targeted therapies.
  • Personalized Treatment Strategies: The established TSRGPM and identified target genes like CDC25A provide a foundation for developing personalized treatment strategies and identifying novel therapeutic targets for gastric cancer.

Summary updated

Abstract

Gastric cancer presents a formidable challenge, marked by its debilitating nature and often dire prognosis. Emerging evidence underscores the pivotal role of tumor stem cells in exacerbating treatment resistance and fueling disease recurrence in gastric cancer. Thus, the identification of genes contributing to tumor stemness assumes paramount importance. Employing a comprehensive approach encompassing ssGSEA, WGCNA, and various machine learning algorithms, this study endeavors to delineate tumor stemness key genes (TSKGs). Subsequently, these genes were harnessed to construct a prognostic model, termed the Tumor Stemness Risk Genes Prognostic Model (TSRGPM). Through PCA, Cox regression analysis and ROC curve analysis, the efficacy of Tumor Stemness Risk Scores (TSRS) in stratifying patient risk profiles was underscored, affirming its ability as an independent prognostic indicator. Notably, the TSRS exhibited a significant correlation with lymph node metastasis in gastric cancer. Furthermore, leveraging algorithms such as CIBERSORT to dissect immune infiltration patterns revealed a notable association between TSRS and monocytes and other cell. Subsequent scrutiny of tumor stemness risk genes (TSRGs) culminated in the identification of CDC25A for detailed investigation. Bioinformatics analyses unveil CDC25A's implication in driving the malignant phenotype of tumors, with a discernible impact on cell proliferation and DNA replication in gastric cancer. Noteworthy validation through <i>in vitro</i> experiments corroborated the bioinformatics findings, elucidating the pivotal role of CDC25A expression in modulating tumor stemness in gastric cancer. In summation, the established and validated TSRGPM holds promise in prognostication and delineation of potential therapeutic targets, thus heralding a pivotal stride towards personalized management of this malignancy.
BibTeX
@inproceedings{li2024machine,
  author = {Guoxing Li and Yunpeng Chen and Youyang Hu and Wenjing Zhao and Yun-Yan Lu and Fujian Wan and Zhijun Wu and Xiangqian Wang and Qiying Yu},
  title = {{Machine learning for identifying tumor stemness genes and developing prognostic model in gastric cancer}},
  booktitle = {Aging},
  year = {2024},
  doi = {10.18632/aging.205715},
}
Figure