Optimizing breast cancer therapy: chemoressitance and machine learning for precision prediction.

Abstract

Background: Validate a novel in vitro resistance platform for breast cancer (BC) by assessing resistance profiles of treatment-naïve and residual tumors after neoadjuvant chemotherapy (NACT) and applying a machine learning algorithm to predict NACT response using clinical biomarkers.

Methods: Tumor cells from primary BC and residual disease (RD) were cultured on the chemoresistance platform with various chemotherapies. Resistance was categorized as low ( < 40%), medium (40-60%), or high ( > 60%) after 72 h based on cell viability. Clinicopathological data from BC samples were analyzed using the XGBoost algorithm and SHAP interpretation to identify NACT-resistant patients.

Results: Patients undergoing upfront surgery (n = 70) exhibited significantly favorable prognosis compared to RD cases (n = 27), which had higher drug resistance and worse outcomes. AI analysis of 1,012 patients achieved 82% accuracy in predicting pathological response and RD, with age, estrogen receptor status, tumor grade and size, axillary status, and HER2 status identified as key predictors. The algorithm predicted NACT resistance with 81.8% accuracy in 11 patient samples.

Conclusion: The chemoresistance platform identified resistance patterns highlighting its utility in precision medicine. Additionally, the XGBoost algorithm accurately predicted NACT response, supporting the integration of AI with functional precision medicine for personalized BC treatment.