Heterogeneous Dimensional Learning Search Algorithm guided by interaction effects for gene selection

• Published in: Journal of computational design and engineering Vol. 12; no. 6; pp. 1 - 41
• Main Authors: Qu, Chiwen, Dong, Shanshan, Qin, Chunlin, Lu, Zenghui
• Format: Journal Article
• Language: English
• Published: Oxford Oxford University Press 01.06.2025; 한국CDE학회
• Subjects: Accuracy; Adaptive control; Algorithms; Cancer; Convergence; Feature recognition; Heuristic methods; Machine learning; Prognosis; Search algorithms; Search methods; 기계공학; learning search algorithm; heterogeneous dimensionality; gene feature selection; interaction effects
• ISSN: 2288-5048; 2288-4300; 2288-5048
• DOI: 10.1093/jcde/qwaf050

Abstract In this paper, we introduce a novel algorithm called the Heterogeneous Dimensional Learning Search Algorithm guided by interaction effects (HDIELSA) for gene feature recognition. HDIELSA combines an interactive search strategy with a heterogeneous dimensional gain factor to boost the algorithm's diversity and search accuracy. By using dynamically adaptive control parameters during both global exploration and local exploitation phases, HDIELSA effectively enhances its global search ability and convergence speed. We conducted extensive experiments on 52 test sets (CEC 2017, CEC 2020, and CEC 2022) and compared HDIELSA with 10 original heuristic algorithms, 6 latest improved algorithms, and the original Learning Search Algorithm. The results, verified through Wilcoxon signed-rank and Friedman tests, show that HDIELSA outperforms other optimization algorithms in terms of convergence speed and accuracy. Additionally, HDIELSA demonstrated excellent performance in evaluating four public microarray expression data sets. In the identification of key genes for glioma prognosis, HDIELSA achieved an accuracy of 87.2%, an Area Under the Curve value of 0.927, and precision, recall, and Receiver Operating Characteristic(ROC) all exceeded 87%. This study not only confirms the effectiveness of the HDIELSA algorithm in selecting cancer gene features but also provides new strategies for early cancer diagnosis and prognosis. Graphical Abstract Graphical Abstract