COMPASS TCGA Dataset

This record provides preprocessed TCGA transcriptomic datasets in pickle (.pkl) format for COMPASS model pretraining. Two versions are included:• TCGA-2475 (subset), containing expression profiles restricted to 2475 immune-related genes curated from hallmark and immune-signature collections.• TCGA-15672, comprising 15,672 protein-coding genes representing a broader transcriptomic space.Each version includes both TPM matrices and patient metadata files, packaged in a directory structure identical to that used in the COMPASS training pipeline to ensure full reproducibility.The datasets were generated using the preprocessing code available at:🔗 https://github.com/mims-harvard/COMPASS-web/tree/main/TCGA_dataset_processingwith guidance and schema documentation described in the COMPASS help page:🔗 https://www.immuno-compass.com/help/index.html#datasets