Abstract

Computational pathology (CompPath) workflows increasingly rely on computational embeddings from digitized hema- toxylin and eosin (H&E) whole-slide images (WSIs). Generating such embeddings reproducibly and at scale requires large image transfers, magnification-aware tiling, model-specific preprocessing, versioned software environments, and substantial graphics processing unit (GPU) inference time. Here, we introduce the Tissue and Report Atlas of Computational Embeddings (TRACE), as a comprehensive atlas of standardized, multi-scale, and multi-encoder pathology embeddings to eliminate the repetition of this inference burden and expedite CompPath discoveries. TRACE provides multi-magnification (5×, 10×, 20×) histopathology embeddings from diagnostic H&E WSIs, as well as text embeddings from associated diagnostic clinical reports, at the scale of The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) entire data collections. In favor of promoting reproducible reuse of these embeddings and clinically-relevant analyses, TRACE also offers patient-level related clinical (e.g., demographic, treatment), molecular, computational (e.g., preprocessing parameters, tensor metadata), and model provenance information. Sanity-check computational validation of the provided embeddings confirm disease-relevant signal, enabling researchers to move directly to expedited, fair, and reproducible CompPath studies. TRACE is released with its accompanying code and documentation via HuggingFace: https://huggingface.co/datasets/IUCompPath/TRACE