Artificial Intelligence to unveil genetic characteristics in Cancer and rare disease
Loading...
Files
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Unuversity Setif 1 Ferhat Abbas . Faculty of Sciences
Abstract
Gene expression analysis has emerged as a powerful approach for cancer biomarker discovery, enabling the identification of molecular signatures associated with tumour initiation, progression, and metastasis. However, the inherent high dimensionality of transcriptomic data, where the number of measured genes far exceeds the number of available patient samples, poses significant challenges for conventional machine learning approaches.
To address these challenges, this thesis proposes an integrated computational framework for gene expression-based biomarker discovery in cancer, applied to the TCGA Pan-Cancer dataset comprising 10,535 samples distributed across 33 cancer types. The proposed framework combines five bio-inspired metaheuristic feature selection algorithms the Bat Algorithm (BA), Differential Evolution (DE), Genetic Algorithm (GA), Grey Wolf Optimizer (GWO), and Particle Swarm Optimization (PSO) with a biological pathway integration strategy leveraging the KEGG, WikiPathways, and STRING databases. The biologically grounded gene subsets are evaluated through a comparative classification analysis using KNN and SVM classifiers, and validated through multi-platform enrichment analysis using GO, ShinyGO, cBioPortal, and STRING.
The experimental results demonstrate that all five algorithms outperformed the baseline classification accuracy of 42.80%, with GWO achieving the best performance by selecting only 371 genes while attaining a KNN accuracy of 74.6%. The pathway integration step reinforced the biological relevance of the identified gene panels, yielding a validated set of 23 candidate biomarkers grounded in well-established cancer signalling mechanisms.
