Baseline immune transcriptional signatures predict checkpoint blockade response in triple-negative breast cancer

Immune checkpoint inhibitors improve clinical outcomes in a subset of patients with triple-negative breast cancer (TNBC), but biomarkers that identify likely responders before treatment remain limited. We asked whether baseline transcriptional profiles from different immune-cell compartments contain multigene information associated with treatment response. We analyzed pre-treatment single-cell RNA sequencing data from 45 patients with TNBC and generated patient-level profiles for T cells, B cells, myeloid cells, and an integrated immune compartment. Random Forest feature selection identified response-associated signatures that were evaluated using logistic regression, feedforward neural networks, and support vector machines. The integrated immune signature showed the strongest development-cohort performance, reaching an accuracy of 0.8936 and an AUC of 0.9335 with support vector machine classification. RF-selected genes were enriched for nominal response-associated expression differences, although no individual gene remained significant after false-discovery-rate correction. CAMERA analysis showed higher interferon-alpha response in responders across all four immune profiles, whereas several metabolic and stress-related Hallmark signatures were higher in non-responders in T-cell and integrated profiles. These functional differences were only partially represented among RF-selected genes. Several reanalyzed T-cell RF features also showed nominal associations with overall survival in TNBC. %\textcolor{red}{} Evaluation in an independent nine-patient TNBC cohort showed heterogeneous performance across immune profiles and classifiers, with discordance between threshold-dependent metrics and AUC in several models. Together, these findings show that baseline immune transcriptional profiles contain multigene information associated with checkpoint-blockade response in TNBC and that integrating T-cell, B-cell, and myeloid transcriptional information improves response discrimination.