Cancer Cachexia Atlas

A browsable collection of published cancer-cachexia data from human and mouse skeletal muscle: 14 datasets spanning bulk RNA-seq, microarray, single-cell and single-nucleus RNA (105,161 cells and nuclei), and mass-spectrometry proteomics. Search a gene, see it across cell types, sample groups or protein groups, and open a full explorer for differential analysis and pathway enrichment.

• Pick a layer (RNA or Protein), then an organism, then scroll the datasets.
• Use the Filters (sidebar, top) to narrow by technology, cancer type or model system.
• Use Jump to dataset (sidebar, below) to scroll to a study.
• Each dataset shows a lightweight inline preview; open the full explorer for interactive analyses.

• RNA ↔ Protein compares transcript and protein effects for the same model, and only for comparisons declared comparable in advance. There are currently 3 such comparisons across 7 contrasts. Each one leads with a coherence test, so a pair of layers that does not track each other says so instead of serving a ranked table.
• Gene across datasets does the opposite on purpose: it puts one feature across all 14 datasets on a single screen, including datasets that are not comparable to each other, because that is the only way to ask whether a gene behaves the same way in different cachexia models. Magnitudes stay on each dataset's own scale, and a dataset that never measured the feature still occupies a row so the denominator is visible.

• Human and Mouse datasets; the structure accommodates clinical siblings.
• Every dataset links out to its repository record — GEO for transcriptomics, ProteomeXchange for proteomics — and to its publication (DOI). Descriptions are the atlas author's own summaries, not reproduced text.
• Some datasets carry an Exploratory caveat where sample size makes contrasts hypothesis-generating — read it before interpreting.
• Where a study's published statistics were unusable, the atlas computes its own and labels them. This applies to 2 of the 3 proteomics datasets, each of which carries a statistics source chip reading published, FDR added by atlas, or computed by atlas.

This atlas presents published data for browsing, and re-derives statistics only where the published ones could not be used — always labelled, never silently. No two datasets here were measured on the same samples, so every cross-dataset comparison is hypothesis-generating rather than a multi-omic measurement.