Integrating single-cell transcriptomics and machine learning to predict breast cancer prognosis and immunotherapy sensitivity: a study based on proliferating T cells

Breast cancer, a highly prevalent and heterogeneous malignancy, poses challenges in prognosis and treatment. Emerging evidence links proliferative T cells (Tprolif) to tumor development and progression. This study aims to enhance prognostic accuracy in breast cancer by identifying Tprolif-associated risk genes for molecular subtyping and developing a prognostic model grounded in these subtypes. Single-cell RNA sequencing (scRNA-seq) data from breast cancer samples were integrated with bulk RNA-seq data from the TCGA-BRCA and GSE20685 cohorts to define Tprolif-based molecular subtypes. A prognostic risk signature was constructed using 10 classical machine learning algorithms, with the optimal algorithm selected through performance testing and validated across multiple datasets. The expression of key model genes was analyzed, and their impact on breast cancer cell proliferation and migration was examined through in vitro experiments. Three distinct molecular subtypes were identified, each characterized by unique immune cell infiltration profiles and survival outcomes. Using Tprolif-based subtyping and machine learning, the random survival forest (RSF) algorithm exhibited superior predictive performance, as measured by the C index. Patients with lower risk scores demonstrated a more active immune microenvironment with greater immune cell infiltration. Functional assays revealed that knockdown of MTFR1 and SLC52A2 markedly inhibited breast cancer cell proliferation and migration. This study underscores the pivotal role of Tprolif cells in breast cancer immune heterogeneity and introduces a robust prognostic model for patient stratification. The identification of molecular subtypes and validation of MTFR1 and SLC52A2 as potential therapeutic targets provide crucial insights for personalized cancer treatment strategies.