Non-invasive prediction of breast cancer biomarkers from blood inflammatory indices using a hybrid adaptive grey wolf optimizer-based machine learning framework

Accurate determination of breast cancer biomarkers, including estrogen receptor (ER), progesterone receptor (PR), HER2/neu, and Ki-67, is essential for treatment planning but typically requires invasive tissue-based assays. Blood-derived inflammatory indices, such as neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) provide a minimally invasive and accessible source of biological information. In this exploratory single-center study, we propose a hybrid Adaptive Grey Wolf Optimizer with Local Search (AGWO-LS)-based machine learning framework for preoperative prediction of biomarker status using routine hematological indices. Data from 151 patients with newly diagnosed breast cancer were analyzed, with age, absolute neutrophil, lymphocyte, and platelet counts, NLR, and PLR as predictor variables. Six classifiers (SVM, k-Nearest Neighbors, Decision Tree, Linear Discriminant, Ensemble, and Random Forest (RF)) were trained and optimized using AGWO-LS, with model evaluation performed via repeated Monte Carlo cross-validation, representing internal validation only. Ki-67 prediction yielded the strongest performance (RF and SVM: 95.6% accuracy; 97.6% F1-score), though these results should be interpreted cautiously given the modest sample size and internal-only validation, which may produce optimistic estimates. ER prediction achieved the second-highest performance (RF: 78.2% accuracy; 85.1% F1-score), while PR and HER2/neu showed moderate and limited performance, respectively. In summary, the proposed approach demonstrates potential for biomarker prediction; however, it is intended as a complementary decision-support tool rather than a replacement for standard immunohistochemical assessment. Given that the findings are based on a relatively small, single-center dataset, external multi-center validation is required to confirm generalizability before clinical application. • We proposed a hybrid AGWO-LS framework for non-invasive breast cancer biomarker prediction. • Routine hematological indices (NLR, PLR, age, blood cell counts) enable non-invasive preoperative prediction of breast cancer IHC biomarkers. • We combine global exploration and local search for improved hyperparameter tuning. • The model is intended as a complementary decision-support tool for preoperative triage and resource-limited settings, not as a replacement for pathology.