Development and validation of a highly accurate multigene gene expression biomarker to predict chemotherapy response in primary triple-negative breast cancer.

PURPOSE: Triple-negative breast cancer (TNBC) is an aggressive subtype lacking estrogen and progesterone receptors and HER2 amplification. Representing 10-15% of breast cancer cases, TNBC disproportionately affects Black and pre-menopausal women and is associated with poorer outcomes. With chemotherapy as the primary systemic treatment option, achieving a pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) is a key prognostic factor. However, TNBC biological heterogeneity complicates treatment response prediction. This study aimed to identify transcriptomic biomarkers predictive of NAC response in TNBC patients and evaluate machine-learning models for response classification.

METHODS: We performed transcriptomic profiling on tumors from 234 TNBC patients, divided into training 138 pCR,72 residual disease (RD) and test 9 pCR, 15 RD cohorts. Feature selection was conducted using LASSO regression and Boruta algorithms to identify robust biomarkers. Random forest and support vector machine (SVM) models were trained on the selected and evaluated on the independent test set.

RESULTS: Feature selection identified 21 overlapping biomarkers, including EPHB3, ATP5MJ, USP1, RANBP9, SLC11A2, S100P, PPP1R1A, ZIC1, NDRG2, SMARCA2, H2BC7, STK24, HBB, VPS45, H1, VEGFA, NFIB, ITGA6, RPRD1A, PRKD3, and ENSA, several of which have been implicated in TNBC progression and treatment resistance. In the test set, predictive performance was strong, with area under the curve (AUC) values of 91% for random forest and 89% for SVM.

CONCLUSION: Transcriptomic profiling combined with machine learning provides a promising approach for predicting NAC response in TNBC. The identified biomarkers may inform precision treatment strategies and improve clinical outcomes in this high-risk patient population.