A Novel Adaptive Sparse Deep Feature Selection Method for Enhanced Gene-based Cancer Classification
Résumé
The Adaptive Sparse Deep Feature Selection (ASDFS) method introduces a novel deep learning-based approach to enhance gene-based cancer classification. Designed to address the high dimensionality and complexity of genomic data, ASDFS leverages sparse autoencoders for dimensionality reduction and a Dual-Target Deep Neural Network (DT-DNN) to refine and identify a minimal yet biologically significant subset of genes. The method achieved outstanding classification accuracies of 99.9%, 100%, and 99.8% for ovarian, prostate, and lung cancers, respectively, achieving superior results compared to state-of-the-art techniques, including Principal Component Analysis with Grey Wolf Optimizer (PCA-GWO), Recursive Feature Elimination (RFE), and Minimum Redundancy Maximum Relevance (mRMR)-based hybrid methods. ASDFS was validated on microarray datasets with detailed characteristics: ovarian cancer (15,154 genes, 253 samples), prostate cancer (12,600 genes, 102 samples), and lung cancer (12,533 genes, 181 samples). This demonstrates its robust performance and ability to achieve significant gene reduction. Additionally, pathway enrichment analysis validated the biological relevance of the selected genes, highlighting their roles in critical cancer pathways.
Citer ce document
Accès au document
Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter
Voir l'article sur le site de la revueAuteur(s)
Statistiques
Consultations : 1
Téléchargements : 0