From multi-omics to clinical practice: BC-BioMIXER to predict PCR and guide individualized neoadjuvant chemotherapy for breast cancer.

Abstract

e12551 Background: Early and accurate prediction of pathological complete response (pCR) is crucial for personalizing neoadjuvant chemotherapy (NACT) in invasive breast cancer, yet many high-performing models depend on costly multi-modal data not routinely available. Purpose: To develop and validate BC-BioMIXER, a biologically informed model that transfers multi-omics–derived knowledge to routine clinical data for pre-treatment pCR prediction. Methods: BC-BioMIXER was developed in a multi-modality cohort of 648 patients with invasive breast cancer (T2–4, any N, M0) with transcriptomic, proteomic, MRI, and clinical data. External validation was performed in three independent cohorts (total N=830): one multi-modality cohort, one clinical trial cohort, and one contemporary real-world cohort. All patients received NACT followed by surgery. The framework uses a teacher–student paradigm: a multi-omics teacher learns biologically integrated representations, which are transferred to a student model trained only on routine clinical data. Performance was benchmarked against a multi-modality reference model and assessed across cohorts, receptor-defined subgroups (HER2, HR), and treatment groups (NACT ± immune checkpoint inhibitors [ICI]). Prognostic value was evaluated using distant recurrence-free survival (DRFS), and clinical utility for immunotherapy was explored by comparing DRFS between NACT+ICI and NACT-alone within model-predicted pCR/non-pCR strata. Results: BC-BioMIXER achieved pCR prediction comparable to the multi-modality benchmark (AUC 0.82 vs 0.85; p=0.271) and showed consistent discrimination across validation cohorts (AUC 0.82, 0.81, 0.80; all p<0.001). Model-predicted pCR was associated with improved 3-year DRFS (HR=0.36; 95% CI, 0.20–0.67; p<0.001). In patients treated with NACT+ICI, BC-BioMIXER was numerically superior to PD-L1 alone for pCR prediction (AUC 0.84 vs 0.72; p=0.08). Notably, within the predicted non-pCR subgroup, NACT+ICI was associated with inferior DRFS versus NACT alone (HR=2.70; p=0.032), while no significant difference was observed in the predicted pCR subgroup. Conclusions: BC-BioMIXER transfers multi-omics biological knowledge to a routine-data model for robust pCR prediction before NACT. Its consistent external validation and ability to stratify outcomes under NACT±ICI support scalable, accessible precision oncology.