Tissue identity check
Is this sample the tissue you think it is?
Drop in an expression table and get back, for every sample, a tissue call, its 90 % prediction set (the tissues the model cannot rule out), the genes behind the call, and a flag when your label doesn't fit.
How it works
- Drop in an expression table. CSV or TSV, the 20-gene panel or a full raw-count matrix.
- Get a call, a 90 % prediction set and the genes behind it, with flags for swapped labels, contamination and unknown tissue.
- Samples from another lab: expect many “Unknown” at first; recalibrate on at least nine of your own labelled samples.
A conformal prediction set lists the tissues the model cannot rule out. On samples like its held-out calibration animals it contains the true tissue at least 90 % of the time; that 90 % describes the sets, not the confidence of any single call. One tissue is a confident call, several is ambiguous, none means the model abstains.
How accurate is it?
What you can upload
- Rat bulk RNA-seq.
- CSV or TSV with a header row.
- Format A: the 20 panel genes in log2 CPM, samples in rows and genes as columns; or format B: a full raw-count matrix, genes × samples or samples × genes, Ensembl IDs or gene symbols (converted to log2 CPM here with the pipeline's transform). Ensembl IDs must be rat rn6 (Rnor_6.0, the build of MoTrPAC release c1.0); with another build, use gene symbols.
- If no tissue clears the calibrated threshold, the sample is marked Unknown: the model abstains.
Optional · flags swaps
A claimed_tissue column (or tissue, true_tissue) with the tissue you think each sample is; it is compared with the call.
Use the tissue names in the list of accepted names (the 19 reference tissues and their synonyms, e.g. brain, muscle, fat).
Show example input tables
A · 20-gene panel (log2 CPM; samples in rows, genes as columns)
B · Full count matrix (raw counts; either orientation; Ensembl IDs shown, gene symbols work too)
Check your samples
Drop a CSV or TSV file here (one file at a time)
Processed locally in your browser. No data leaves this page.