Scholarlink

Background <p>Myelodysplastic Syndrome (MDS) comprises a heterogeneous group of clonal hematopoietic stem cell disorders characterized by ineffective hematopoiesis, peripheral blood cytopenias, and an increased risk of progression to acute myeloid leukemia (AML). Despite extensive research into the molecular pathogenesis of MDS, there remains a critical need for reliable diagnostic biomarkers and therapeutic targets to improve clinical management.</p> Objective <p>The present study aims to identify robust gene expression signatures and key regulatory pathways associated with MDS through integrative transcriptomic analysis, with the goal of enhancing diagnostic accuracy and informing biomarker discovery.</p> Methods <p>An integrative bioinformatics approach was employed using transcriptomic data from four publicly available microarray datasets (GSE4619, GSE19429, GSE30195, and GSE58831), encompassing a total of 461 samples. Differentially expressed genes (DEGs) were identified, followed by functional enrichment analysis to elucidate disrupted biological processes. Protein–protein interaction (PPI) networks were constructed to assess gene connectivity, and hub genes were prioritized via Maximal Clique Centrality (MCC). A panel of 20 hub genes was subsequently used to train multiple supervised learning models. Among them, a Support Vector Machine (SVM) classifier with a Radial Basis Function (RBF) kernel demonstrated superior diagnostic performance. Model generalizability was assessed using two independent external datasets (GSE114922 and GSE2779).</p> Results <p>A total of 543 DEGs were identified, comprising 320 upregulated and 223 downregulated genes. Functional enrichment revealed significant perturbations in erythropoiesis, immune-related pathways, and transcriptional regulation. PPI network analysis revealed 20 highly connected hub genes, enriched with interferon signaling and B-cell lineage functions. The SVM-RBF classifier achieved an accuracy of 99.39% and an AUC of 0.9998. External validation confirmed the model’s robustness, yielding 100% sensitivity and over 91% accuracy in both validation cohorts.</p> Conclusion <p>This study presents a comprehensive integrative framework for the transcriptomic classification of MDS, highlighting key gene signatures with high diagnostic potential. The identified hub genes represent promising candidates for the future development of targeted molecular diagnostics and may provide insights into MDS pathobiology, supporting advancements in precision hematology.</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

Multicohort transcriptomic integration and machine learning-based diagnostic modeling for myelodysplastic syndromes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月7日

查看全文
  • Rana Hossam Elden,
  • Nancy M. Salem

摘要

Background

Myelodysplastic Syndrome (MDS) comprises a heterogeneous group of clonal hematopoietic stem cell disorders characterized by ineffective hematopoiesis, peripheral blood cytopenias, and an increased risk of progression to acute myeloid leukemia (AML). Despite extensive research into the molecular pathogenesis of MDS, there remains a critical need for reliable diagnostic biomarkers and therapeutic targets to improve clinical management.

Objective

The present study aims to identify robust gene expression signatures and key regulatory pathways associated with MDS through integrative transcriptomic analysis, with the goal of enhancing diagnostic accuracy and informing biomarker discovery.

Methods

An integrative bioinformatics approach was employed using transcriptomic data from four publicly available microarray datasets (GSE4619, GSE19429, GSE30195, and GSE58831), encompassing a total of 461 samples. Differentially expressed genes (DEGs) were identified, followed by functional enrichment analysis to elucidate disrupted biological processes. Protein–protein interaction (PPI) networks were constructed to assess gene connectivity, and hub genes were prioritized via Maximal Clique Centrality (MCC). A panel of 20 hub genes was subsequently used to train multiple supervised learning models. Among them, a Support Vector Machine (SVM) classifier with a Radial Basis Function (RBF) kernel demonstrated superior diagnostic performance. Model generalizability was assessed using two independent external datasets (GSE114922 and GSE2779).

Results

A total of 543 DEGs were identified, comprising 320 upregulated and 223 downregulated genes. Functional enrichment revealed significant perturbations in erythropoiesis, immune-related pathways, and transcriptional regulation. PPI network analysis revealed 20 highly connected hub genes, enriched with interferon signaling and B-cell lineage functions. The SVM-RBF classifier achieved an accuracy of 99.39% and an AUC of 0.9998. External validation confirmed the model’s robustness, yielding 100% sensitivity and over 91% accuracy in both validation cohorts.

Conclusion

This study presents a comprehensive integrative framework for the transcriptomic classification of MDS, highlighting key gene signatures with high diagnostic potential. The identified hub genes represent promising candidates for the future development of targeted molecular diagnostics and may provide insights into MDS pathobiology, supporting advancements in precision hematology.