Abstract
Breast cancer is highly prevalent and fatal worldwide. Currently, breast cancer classification is based on the presence of estrogen, progesterone, and human epidermal growth factor 2. Because cancer and metabolism are closely related, we established a breast cancer classification system based on the metabolic gene expression profile. We performed typing of metabolism-related genes using The Cancer Genome Atlas-Breast Cancer and 2010 (YAU). We included 2,752 metabolic genes reported in previous literature, and the genes were further identified according to statistically significant variance and univariate Cox analyses. These prognostic metabolic genes were used for non-negative matrix factorization (NMF) clustering. Then, we identified characteristic genes in each metabolic subtype using differential analysis. The top 30 characteristic genes in each subtype were selected for signature construction based on statistical parameters. We attempted to identify standard metabolic signatures that could be used for other cohorts for metabolic typing. Subsequently, to demonstrate the effectiveness of the 90 Signature, NTP and NMF dimensional-reduction clustering were used to analyze these results. The reliability of the 90 Signature was verified by comparing the results of the two-dimensionality reduction clusters. Finally, the submap method was used to determine that the C1 metabolic subtype group was sensitive to immunotherapy and more sensitive to the targeted drug sunitinib. This study provides a theoretical basis for diagnosing and treating breast cancer.