Abstract B054: Identification of spatial motifs linked to tumor genotype using graph attention networks
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Abstract Spatial analysis of cancer requires computational methods capable of addressing the complexity and heterogeneity characteristic of tumor tissue architectures. Unlike well-organized tissues such as developing organs or the brain, cancer tissues lack easily identifiable spatial motifs, necessitating computational models that model a principled definition of a spatial motif tailored to their unique structure. We present GRAFITI, a graph autoencoder specifically designed to identify spatial motifs within cancer tissues using a formal graph-based definition. GRAFITI employs state of the art machine learning methods, including a deep multi-head graph-attention (GAT) encoder with dual decoders that reconstruct both expression profiles and spatial relationships among cells. The latent representation produced by GRAFITI is clustering using an integrated clustering head, that refines the spatial motif annotations during training. Using this output, GRAFITI minimizes an objective function designed around intra-motif similarity, inter-motif dissimilarity, and explicit spatial coherence constraints. GRAFITI employs additional spatial regularization techniques—such as continuity and separation losses—to effectively manage the spatial noise typical of disorganized cancer tissues. Its attention mechanism enhances interpretability by dynamically highlighting significant cell-cell interactions, such as tumor-immune infiltration events, identifying meaningful relationships within chaotic spatial structures. We validated GRAFITI using semi-synthetic datasets specifically designed to replicate common cancer tissue features, demonstrating superior performance in identifying spatial motifs compared to current models, as measured by the adjusted Rand index (ARI). Given that cancer is fundamentally driven by genomic alterations, aligning spatial organization to underlying genomic features has potential for deeper understanding of spatial organization in tumor biology. To this end, GRAFITI has been applied to imaging datasets from ovarian, melanoma, and breast cancers, where the distribution of learned spatial motifs across images can be mapped to genomic features. Citation Format: Nicholas Ceglia, Maryam Pourmaleki, Alessandro Grande, Adam Weiner, Jose Meza Llamosas, Andrew McPherson, Sohrab Shah. Identification of spatial motifs linked to tumor genotype using graph attention networks [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 B054.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.006 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it