Abstract Background: TNBC remains a biologically aggressive subtype with limited biomarkers to predict pathologic response in the neoadjuvant setting. While spatial transcriptomics (ST) can reveal features of the tumor microenvironment (TME) associated with response and resistance, its clinical application is limited by cost and scalability. As part of ongoing efforts to investigate mechanisms of resistance to neoadjuvant chemotherapy and ICI in breast cancer, we evaluated the feasibility of virtual RNA inference (VRI)—a deep learning approach for prediction of spatial gene expression from routine H 0.001), G2M checkpoint (p 0.001), and epithelial-mesenchymal transition (p = 0.03), among others. A composite score derived from these pathways effectively stratified responders from non-responders, achieving an AUC of 0.82 (sensitivity=0.77, specificity=0.73). Ongoing works include expansion of include cohort size, additional subgroup analysis to factor in treatment regimens (e.g., ICI), and adjust for clinical covariates such as stage, grade, residual tumor, and tumor cellularity. Conclusion: This pilot study demonstrates the feasibility of using VRI to infer tumor-specific gene expression patterns predictive of neoadjuvant chemotherapy response from standard H 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS2-09-05.
Yuan et al. (Tue,) studied this question.