Abstract B045: Deep learning–guided spatial dissection of melanoma uncovers compartmentalized tumor states associated with response and resistance to immunotherapy and its combination with MAPK inhibitors
Bibliographic record
Abstract
Abstract Purpose: Intratumoral heterogeneity is a key determinant of immunotherapy resistance, yet current biomarkers largely rely on single-modality features—such as CD8+ T cell infiltration, IFNG expression, or checkpoint ligand staining—without accounting for the spatial complexity of the tumor microenvironment. Composite biomarkers that integrate both tumor-intrinsic programs and their spatial organization remain underdeveloped. To address this, we combined machine learning–based molecular classification with spatial transcriptomics to uncover how transcriptional tumor states are physically structured within the tumor architecture and how this organization governs response or resistance to immunotherapy. Methods: We leveraged non-negative matrix factorization (NMF) and tumor microenvirnoment metabolism signatures with bulk RNA-seq data from >700 melanoma tumors across multiple phase I–III clinical trials. These programs were mapped onto spatial transcriptomics data using Tangram 2.0, a deep learning model that aligns single-cell or deconvolved bulk data to spatial coordinates. We analyzed cell–cell interactions, tumor-stromal patterning, and radial gene gradients to interpret functional spatial architecture. Results: Spatial mapping revealed that melanoma tumors often harbor multiple transcriptional programs arranged in discrete, non-overlapping tissue regions. Notably, undifferentiated (UR) and differentiated (DC) programs—associated with distinct therapeutic responses—co-existed within the same tumor but remained compartmentalized. Only spatial domains with DC-like features exhibited MAPK inhibitor–induced MHC-I expression and immune cell infiltration, whereas UR regions remained immune-excluded and were enriched for fibroblasts and extracellular matrix remodeling. These findings suggest that spatial segregation of tumor states, rather than overall tumor composition, may predict response to immunotherapy. Conclusions: This study presents a machine learning–enabled spatial framework that links transcriptional identity to physical tissue structure in melanoma. By uncovering how spatial architecture constrains therapeutic response, our work highlights the value of AI-driven models like Tangram for interpreting spatial data and guiding precision immuno-oncology. Citation Format: Kalpit Shah. Deep learning–guided spatial dissection of melanoma uncovers compartmentalized tumor states associated with response and resistance to immunotherapy and its combination with MAPK inhibitors [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B045.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".