A tumour's molecular profile is dominated by the organ it came from, so a model can look impressive by recognising the organ and must be judged against a baseline that knows only that.
The TCGA Pan-Cancer Atlas classification paper found that cell of origin dominates the molecular clustering of ten thousand tumours across thirty-three types. Expression, methylation and even mutation patterns encode the tissue so strongly that a pan-cancer survival or drug-response model gets much of its apparent accuracy from the tissue label; the honest comparison is a tissue-only or cancer-type-only baseline, and within-type (stratified) metrics. The same signal is what classifiers exploit to assign a primary site to cancers of unknown primary (Wikipedia).
Shares Accuracy, macro-F1 and confusion matrices, C-index (concordance index), Harrell's and Uno's, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Gene co-expression structure, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Accuracy, macro-F1 and confusion matrices, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Gene co-expression structure, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares C-index (concordance index), Harrell's and Uno's, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Clinical covariates (age, stage, nodes, treatment flags), Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Clinical covariates (age, stage, nodes, treatment flags), Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares C-index (concordance index), Harrell's and Uno's, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.