MORSA: Processed Results and Analysis Source Data for Morphology-Omics Representation via Structural Alignment
Year: 2026
Type: dataset
Abstract: This record provides the processed results and analysis source data supporting MORSA (Morphology-Omics Representation via Structural Alignment), a computational framework for predicting transcriptomic coordinates from hematoxylin and eosin whole-slide images through explicit alignment with RNA-derived subspace structure. The dataset covers 7,339 evaluated TCGA WSI-RNA records across 16 cancer types and 1,253 external records across ten cohort-cancer pairs. It contains processed observations supporting transcriptome-recovery evaluation, structural analyses, gene-level biological interpretation, analytical-model experiments, downstream molecular readouts, and virtual cartography. The archive also includes patient-level cross-validation assignments, experiment definitions, fold-level runtime measurements, gene-wise recovery metrics, RNA and morphology spectra, LearnedRank-PCA subspace comparisons, rank-sensitivity results, enrichment analyses, molecular-program metrics, and tile-level program scores. The released results correspond to the fixed experimental configuration using training seed 29, patient split seed 0, RNA basis seed 29, five-fold patient-level cross-validation, and selected rank K=16. Candidate ranks K={8, 16, 32, 48, 64} were evaluated using held-out validation predictions. The archive includes a README, data schema, experiment configuration, provenance information, third-party data notes, and SHA256 checksums. Raw whole-slide images, raw RNA-seq files, UNI feature arrays, the gated UNI checkpoint, and complete trained model checkpoints are not redistributed. Public source identifiers and access routes are provided where applicable. Code and reproducibility workflows:https://github.com/geogr0n/MORSA Dataset release:v1.1.1 Current version DOI:10.5281/zenodo.21436204 Concept DOI for all versions:10.5281/zenodo.21195955 Frozen code release:v1.0.0 Code tag:v1.0.0 Code commit:904a5ddc9e3ca68f2d2f45cafe9f41469549c199 Version 1.1.1 corrects the release metadata in the archive so that the dataset DOI and frozen public code commit match the published records. Scientific result files and reported numerical values are unchanged. These data are associated with a manuscript in preparation. Author-generated processed files are released under the Creative Commons Attribution 4.0 International license. Third-party resources remain governed by their original access and licensing terms.
Author: Chen, Shuo
Institution: Changchun University of Science and Technology
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7167412571
