A novel feature selection algorithm for identifying hub genes in lung cancer

Abstract

Lung cancer, a life-threatening disease primarily affecting lung tissue, remains a significant contributor to mortality in both developed and developing nations. Accurate biomarker identification is imperative for effective cancer diagnosis and therapeutic strategies. This study introduces the Voting-Based Enhanced Binary Ebola Optimization Search Algorithm (VBEOSA), an innovative ensemble-based approach combining binary optimization and the Ebola optimization search algorithm. VBEOSA harnesses the collective power of the state-of-the-art classification models through soft voting. Moreover, our research applies VBEOSA to an extensive lung cancer gene expression dataset obtained from TCGA, following essential preprocessing steps including outlier detection and removal, data normalization, and filtration. VBEOSA aids in feature selection, leading to the discovery of key hub genes closely associated with lung cancer, validated through comprehensive protein-protein interaction analysis. Notably, our investigation reveals ten significant hub genes-ADRB2, ACTB, ARRB2, GNGT2, ADRB1, ACTG1, ACACA, ATP5A1, ADCY9, and ADRA1B-each demonstrating substantial involvement in the domain of lung cancer. Furthermore, our pathway analysis sheds light on the prominence of strategic pathways such as salivary secretion and the calcium signaling pathway, providing invaluable insights into the intricate molecular mechanisms underpinning lung cancer. We also utilize the weighted gene co-expression network analysis (WGCNA) method to identify gene modules exhibiting strong correlations with clinical attributes associated with lung cancer. Our findings underscore the efficacy of VBEOSA in feature selection and offer profound insights into the multifaceted molecular landscape of lung cancer. Finally, we are confident that this research has the potential to improve diagnostic capabilities and further enrich our understanding of the disease, thus setting the stage for future advancements in the clinical management of lung cancer. The VBEOSA source codes is publicly available at https://github.com/TEHNAN/VBEOSA-A-Novel-Feature-Selection-Algorithm-for-Identifying-hub-Genes-in-Lung-Cancer .

Figure 1

The proposed methodology.

Algorithm 1

Pseudocode VBEOSA algorithm.

Figure 2

The accuracy of models based on 50 population size.

Figure 3

Complete PPIN of differentially expressed genes in lung cancer.

Figure 4

Zero order PPI.

Figure 5

Top 10 hub genes network.

Figure 6

Dendrogram showing the clustering of DEGs using a dissimilarity measure.

Figure 7

The plot on the left visually displays the clustering of genes through the utilization of dissimilarity measures based on topological overlap (TOM). On the right, the plot illustrates a hierarchical clustering dendrogram, revealing the relationships among module eigengenes. In this representation, nodes are labeled according to their respective module color names, providing insights into the interconnectedness within the eigengene network.

Figure 8

The plot on the left visually presents the outcome of power values in relation to the scale-independence of genes within co-expression modules associated with lung cancer. On the right, the plot showcases the influence of power values on the average connectivity of genes within co-expression modules related to lung cancer.

Figure 9

Correlation Heatmap between clinical attributes and module eigengenes based on lung cancer.