Thu, Jun 18, 2026
[Archive]
Volume 16, Issue 3 (6-2026)
Iran J Ped Hematol Oncol 2026, 16(3): 934-947 |
Back to browse issues page
Ethics code: IR.SSU.MEDICINE.REC.1401.143
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Sheikhpour R, Zangeneh Soroush M, Hashemi A S. Adaptive Structure Learning for Leukemia Biomarker Discovery from Gene Expression Data. Iran J Ped Hematol Oncol 2026; 16 (3) :934-947
URL: http://ijpho.ssu.ac.ir/article-1-1045-en.html
URL: http://ijpho.ssu.ac.ir/article-1-1045-en.html
Department of Computer Engineering, Faculty of Engineering, Ardakan University, P.O. Box 184, Ardakan, Iran & Department of Computer Engineering, Faculty of Engineering, Ardakan University, P.O. Box 184, Ardakan, Iran
Abstract: (23 Views)
Background: Leukemia classification based on gene expression data is a challenging problem due to the high dimensionality of microarray datasets and the limited number of patient samples. Identifying a small subset of informative genes (biomarkers) that can accurately distinguish between acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) is essential for improving diagnostic accuracy and supporting precision medicine approaches.
Materials and Methods: In this methodological study, an adaptive structure learning framework is proposed for biomarker discovery using the Golub Leukemia Dataset. The proposed method jointly integrates adaptive similarity structure learning, sparse feature selection, and sample reweighting into a unified optimization model. This framework learns both the intrinsic geometric structure of the data and the most discriminative gene subset simultaneously. The selected genes were evaluated using two classifiers, including k-nearest neighbor (KNN) and support vector machine (SVM).
Results: Experimental results show that the proposed method achieves 97.06% classification accuracy using KNN with 8 selected genes and 100% accuracy using SVM with only 7 genes. Comparative analysis with existing feature selection methods demonstrates that the proposed approach achieves superior or competitive performance while using a significantly smaller number of genes.s
Conclusion: The proposed framework effectively identifies compact and highly discriminative biomarker sets for leukemia classification. By jointly modeling sample relationships and gene relevance, the method improves classification performance while reducing feature dimensionality. The results suggest that the proposed framework has potential applications in clinical decision-support systems and other high-dimensional biomedical classification problems.
Materials and Methods: In this methodological study, an adaptive structure learning framework is proposed for biomarker discovery using the Golub Leukemia Dataset. The proposed method jointly integrates adaptive similarity structure learning, sparse feature selection, and sample reweighting into a unified optimization model. This framework learns both the intrinsic geometric structure of the data and the most discriminative gene subset simultaneously. The selected genes were evaluated using two classifiers, including k-nearest neighbor (KNN) and support vector machine (SVM).
Results: Experimental results show that the proposed method achieves 97.06% classification accuracy using KNN with 8 selected genes and 100% accuracy using SVM with only 7 genes. Comparative analysis with existing feature selection methods demonstrates that the proposed approach achieves superior or competitive performance while using a significantly smaller number of genes.s
Conclusion: The proposed framework effectively identifies compact and highly discriminative biomarker sets for leukemia classification. By jointly modeling sample relationships and gene relevance, the method improves classification performance while reducing feature dimensionality. The results suggest that the proposed framework has potential applications in clinical decision-support systems and other high-dimensional biomedical classification problems.
Type of Study: Research |
Subject:
Hematology
Received: 2026/05/14 | Accepted: 2026/06/10 | Published: 2026/06/17
Received: 2026/05/14 | Accepted: 2026/06/10 | Published: 2026/06/17
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
