Predicting gene signature in breast cancer patients with multiple machine learning models

The aim of this study was to predict gene signatures in breast cancer patients using multiple machine learning models.

In this study, we first collated and merged the datasets GSE54002 and GSE22820, obtaining a gene expression matrix comprising 16,820 genes (including 593 breast cancer (BC) samples and 26 normal control (NC) samples). Subsequently, we performed enrichment analyses using Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Disease Ontology (DO).

We identified 177 differentially expressed genes (DEGs), including 40 up-regulated and 137 down-regulated genes, through differential expression analysis. The GO enrichment results indicated that these genes are primarily involved in extracellular matrix organization, positive regulation of nervous system development, collagen-containing extracellular matrix, heparin binding, glycosaminoglycan binding, and Wnt protein binding, among others. KEGG enrichment analysis revealed that the DEGs were primarily associated with pathways such as focal adhesion, the PI3K-Akt signaling pathway, and human papillomavirus infection. DO enrichment analysis showed that the DEGs play a significant role in regulating diseases such as intestinal disorders, nephritis, and dermatitis. Further, through LASSO regression analysis and SVM-RFE algorithm analysis, we identified 9 key feature DEGs (CF-DEGs): ANGPTL7, TSHZ2, SDPR, CLCA4, PAMR1, MME, CXCL2, ADAMTS5, and KIT. Additionally, ROC curve analysis demonstrated that these CF-DEGs serve as a reliable diagnostic index. Finally, using the CIBERSORT algorithm, we analyzed the infiltration of immune cells and the associations between CF-DEGs and immune cell infiltration across all samples.

Our findings provide new insights into the molecular functions and metabolic pathways involved in breast cancer, potentially aiding in the discovery of new diagnostic and immunotherapeutic biomarkers.