Machine learning revealed inflammatory features and a novel risk score-based classification with appealing implications in discriminating the prognosis, immunotherapy and chemotherapy

Abstract

Abstract Breast cancer is the most common and ranks the second cause of related cancer-related death in women. However, the relationship between inflammation-related gene signatures and the prognosis of breast cancer remains elusive. We constructed inflammation related gene features to evaluate the prognosis, immunotherapy, inflammatory response and chemotherapy of breast cancer patients, including 4073 breast cancer patients (TCGA: 1104, GEO: 2969). ● Through univariate Cox regression, LASSO, stepwise regression, and multiple Cox regression analyses, we found that the signatures of inflammatory genes (including 12 genes) can be used to classify breast cancer patients. Patients in the higher risk score group had a poorer prognosis, often accompanied by a higher abundance of macrophages and a lower abundance of lymphocytes, suggesting that inflammation was present in high-risk individuals. ● Tumor mutational burden (TMB) and drug sensitivity analysis showed that PD 0332991, ROSCOVITINE has a higher drug sensitivity to the treatment of low-risk inflammatory breast cancer, while it has a higher drug sensitivity to high-risk patients than carlumide and imatinib. ● Based on risk scores and clinical data, we use training sets and test sets to build nomographs that can be used to calculate patient survival. Our study provides not only insights into the identification of breast cancer patients with poor prognoses, but also treatment strategiesfor breast cancer patients.