Raw Data for UniCure: A Multi-Modal Model for Predicting Personalized Cancer Therapy Response
Description
This dataset contains the raw data files used to reproduce the paper-related analyses and figure panels for UniCure, a multi-modal model that integrates cellular omics representations and chemical representations to predict transcriptomic drug responses across cancer-related cellular contexts. The files are organized by figure and analysis module, including model-evaluation summaries, benchmark metrics, training-loss summaries, t-SNE coordinates, target-gene response tables, pathway-enrichment outputs, fine-tuning and ablation results, patient stratification matrices, survival-analysis inputs, drug-recommendation summaries, and natural-product screening/ranking outputs. This dataset is intended to be used together with the UniCure source code and reproducibility scripts available at: https://github.com/ZexiChen502/UniCure The figure-level reproducibility scripts are provided in the Reproducibility/ directory, and the detailed data-to-result workflow is described in: Reproducibility/UniCure_Paper_Results_Reproducibility_Guide.md This raw-data package has also been reviewed using the Scientific Raw Data Audit framework: https://github.com/ZexiChen502/Scientific-Raw-Data-Audit The audit report and the accompanying point-by-point response are included with the dataset and should be read together with the data files. The audit identified contextual yellow flags requiring review, but no red files. The response document explains why the flagged patterns are expected for specific derived outputs, including integer rank matrices, fixed-denominator averages, retained real-output tables, and pathway-analysis result files.