AI-DRIVEN SPATIAL RECONSTRUCTION OF MINIMAL RESIDUAL DISEASE AND RECURRENCE TOPOLOGIES IN TRIPLE-NEGATIVE BREAST CANCER: INTEGRATIVE MODELING USING SPATIAL TRANSCRIPTOMICS AND IMMUNE-METASTATIC INTERACTIONS rifaldyfajar251@gmail.com

 

AI-DRIVEN SPATIAL RECONSTRUCTION OF MINIMAL RESIDUAL DISEASE AND RECURRENCE TOPOLOGIES IN TRIPLE-NEGATIVE BREAST CANCER: INTEGRATIVE MODELING USING SPATIAL TRANSCRIPTOMICS AND IMMUNE-METASTATIC INTERACTIONS

4

Indonesia

Poster Exhibition only

Triple-negative breast cancer (TNBC) has a high recurrence rate due to spatially concealed minimal residual disease (MRD) post-surgery. Current resection planning lacks predictive tools to identify MRD-enriched niches. This study aims to develop an AI-based spatial model to identify residual tumor niches and predict recurrence trajectories by integrating transcriptomic and immune-topologic features.

We integrated datasets GSE299631 (spatiotemporal MRD in mouse/human BRCA1-TNBC), GSE299393 and GSE300613 (spatial and scRNA-seq from MDA-MB-231 xenografts), GSE266919 (immune modulation under chemotherapy/ICB), and GSE294399 (CTC–T cell clustering). Graph neural networks (GNNs) were used to model spatial transcriptomic architecture. Deep manifold alignment captured MRD lineage transitions. Transformer-based sequence modeling reconstructed temporal escape patterns from 423,157 single-cell and 24,586 spatial transcriptomic profiles.

We identified a SOX4⁺/FOSL2⁺/CXCL13⁺ tumor population within MRD clusters adjacent to CX3CR1⁺ macrophages and TOX⁺ PD-1⁺ CD8⁺ T cells, forming an immunosuppressive niche. These zones were enriched in 68.2% of residual regions (95% CI: 60.9–75.5%) and predicted recurrence with AUROC 0.793. Metastatic escape routes inferred from MRD sites matched known liver/lung metastases with 87.5% accuracy. VLA4⁺ DPT cell depletion was significant near MRD (p = 0.002), suggesting impaired immune containment. Spatial coupling scores exceeded 0.83 in CXCL13-rich niches with immune exclusion.

Our model provides a computational framework to identify minimal residual disease (MRD) zones and recurrence risk areas before surgery, supporting more precise resection strategies and potential integration with neoadjuvant immunotherapy. This approach redefines MRD not only as a post-operative histologic finding but as a spatially predicted risk feature that can inform surgical decisions in TNBC.