Topological Tumor Graphs

Topological Tumor Graphs maps spatial relationships in melanoma histology to infer stromal recruitment and assess its contribution to immunosuppression and patient prognosis.

Key Features:

• Spatial Analysis: Performs spatially explicit analyses of hematoxylin and eosin (H&E) stained melanoma specimens to quantify distributions of stromal, immune, and cancer cells.
• Cell Classification: Uses the CRImage computational pathology pipeline to classify cells in H&E specimens as stromal, immune, or cancer cells, validated against tumor purity assessments, pathologists' estimates of lymphocyte density, imputed immune cell subtypes, and pathway analyses.
• Graph-Based Algorithm: Computes spatial interactions between classified cell types via a graph-based algorithm to identify stromal features termed "stromal clustering" and "stromal barrier."
• Prognostic Associations: Associates increased stromal clustering and stromal barrier with reduced intratumoral lymphocyte distribution and poorer overall survival, independent of existing prognostic factors.
• Genomic Integration: Integrates copy number and transcriptomic data using deep learning to infer a compressed representation of copy number-driven alterations in gene expression, revealing reduced expression of naïve CD4, MAPK, and PI3K signaling pathways in tumors with high stromal clustering and barriers.

Scientific Applications:

• Immunotherapy Insights: Elucidates the immunosuppressive role of stromal cells and T-cell exclusion near melanoma cells to inform immunotherapy research.
• Prognostic Biomarker Development: Enables development of prognostic biomarkers in metastatic melanoma by linking stromal phenotypes to patient outcomes.
• Pathway Analysis: Identifies pathways affected by stromal features to clarify molecular mechanisms of immune exclusion and tumor progression.

Methodology:

Performs spatially explicit analysis of H&E specimens from The Cancer Genome Atlas, applies the CRImage computational pathology pipeline for cell classification, computes interactions between cell types using a graph-based algorithm, and integrates copy number and transcriptomic data via deep learning to derive compressed representations of copy number-driven expression alterations.