Identification of candidate biomarkers correlated with the pathogenesis of breast cancer patients.

Breast cancer (BC) is the second leading cause of cancer-related death in females, followed by lung cancer. Disadvantages exist in conventional diagnostic techniques of BC, such as radiation risk. The present study integrated bioinformatics analysis with machine learning to elucidate potential key candidate genes associated with the tumorigenesis of BC. Eleven datasets were downloaded from the Gene Expression Omnibus (GEO) database and were consolidated into two independent cohorts (training cohort and validation cohort) after batch-effect removal. We employed "limma" package to screen differentially expressed genes (DEGs) between BC and adjacent normal breast samples. Subsequently, the most reliable diagnostic indicators were identified utilizing LASSO-Logistic regression, SVM-RFE and multivariate stepwise Logistic regression analysis. Logistic model and nomogram were created based on these hub genes and applied in external validation cohort to verify the robustness of the model. As a result, a total of six hub genes connected with BC pathogenesis were identified, including CD300LG, IGSF10, FAM83D, MAMDC2, COMP and SEMA3G. Then, a diagnostic model of BC on the basis of these genes was established. ROC analysis of the diagnostic model illustrated that AUC of the training cohort was 0.978 (0.962, 0.995). In the validation cohort, AUC of training set and validation set were 0.936 (0.910, 0.961) and 0.921 (0.870, 0.972), respectively. This indicated that the model was reliable in separating BC patients from healthy individuals. The model may assist in early diagnosis of BC with implications for improving the prognosis of BC patients.