A Multiresolution Breast Cancer CIBERSORTx Resource Validated for Accuracy, Interpretive Limits, and Biological and Clinical Coherence in Tumor Microenvironment Deconvolution

Accurate deconvolution of bulk transcriptomes is essential for characterizing the breast cancer tumor microenvironment (TME), yet existing reference matrices incompletely capture tumor-specific cellular diversity. Here, we developed breast cancer–specific multiresolution CIBERSORTx signature matrices from single-cell RNA sequencing data and systematically evaluated their analytical performance and interpretability. Major-, minor-, and subset-level matrices were constructed and assessed using pseudo-bulk mixtures and pure cell profiles, while biological and clinical coherence were evaluated in TCGA-BRCA and the I-SPY2 cohort. All matrices demonstrated high accuracy in reconstructing pseudo-bulk compositions, with performance declining at finer resolution. Spillover increased with granularity but was largely restricted within related lineages. Lineage-wise deconvolution modestly reduced spillover but consistently decreased accuracy, highlighting the importance of cross-lineage transcriptional contrast. In external datasets, most inferred cell populations showed biologically coherent associations with canonical markers and pathways, whereas some fine-resolution subsets exhibited non-canonical patterns, likely reflecting intra-lineage trade-offs or context-dependent transcriptional states. In the I-SPY2 cohort, plasmablasts and selected myeloid populations were positively associated with pathological complete response, whereas fibroblastic and perivascular-like populations showed negative associations. These findings establish a validated and interpretable resource for breast cancer TME deconvolution and clarify its performance characteristics and limitations.