Triple negative breast cancer (TNBC) is the most aggressive breast cancer (BC) and often affects young women. TNBCs are highly heterogeneous and do not benefit from personalized medicine at localized stages. Most TNBC patients undergo neoadjuvant chemotherapy (NAC) before surgery. In case of chemoresistance with residual tumor after NAC, survival is poor despite execution of complete tumor resection. There is currently no clinically useful biomarker to predict TNBC chemoresistance to NAC that would enable targeted therapeutic intensification. We analyzed here a unique cohort of 106 TNBC tumors before NAC, including 58 chemoresistant and 48 chemosensitive cases, from 2 independent hospitals. Using machine learning under a nested cross-validation design, we obtained two transcriptomic signatures respectively generated from standard differential gene expression analysis and reference-free analysis of differential fragments of transcripts, without any annotation bias. This approach resulted in accurate signatures of TNBC chemoresistance to NAC. Gene ontology analyses of reference-free signatures highlighted DNA repair, replication, and metabolism, in agreement with current knowledge of TNBC resistance biology. In summary, these results show the potential of a reference-free generated transcriptomic signature as predictive biomarker of early TNBC chemoresistance.
© The Author(s) 2025. Published by Oxford University Press.