Machine Learning-Based Prognostic Signature in Breast Cancer: Regulatory T Cells, Stemness, and Deep Learning for Synergistic Drug Discovery

Samina Gul, Jianyu Pang, Yongzhi Chen, Qi Qi, Yuheng Tang, Yingjie Sun, Hui Wang, Wenru Tang, Xuhong Zhou

Regulatory T cells (Tregs) have multiple roles in the tumor microenvironment (TME), which maintain a balance between autoimmunity and immunosuppression. This research aimed to investigate the interaction between cancer stemness and Regulatory T cells (Tregs) in the breast cancer tumor immune microenvironment. Breast cancer stemness was calculated using one-class logistic regression. Twelve main cell clusters were identified, and the subsequent three subsets of Regulatory T cells with different differentiation states were identified as being closely related to immune regulation and metabolic pathways. A prognostic risk model including MEA1, MTFP1, PASK, PSENEN, PSME2, RCC2, and SH2D2A was generated through the intersection between Regulatory T cell differentiation-related genes and stemness-related genes using LASSO and univariate Cox regression. The patient’s total survival times were predicted and validated with AUC of 0.96 and 0.831 in both training and validation sets, respectively; the immunotherapeutic predication efficacy of prognostic signature was confirmed in four ICI RNA-Seq cohorts. Seven drugs, including Ethinyl Estradiol, Epigallocatechin gallate, Cyclosporine, Gentamicin, Doxorubicin, Ivermectin, and Dronabinol for prognostic signature, were screened through molecular docking and found a synergistic effect among drugs with deep learning. Our prognostic signature potentially paves the way for overcoming immune resistance, and blocking the interaction between cancer stemness and Tregs may be a new approach in the treatment of breast cancer.