Fig. 3

Confusion matrices showing the performance, in terms of accuracy (ACC), balanced accuracy (BACC) and AUC score, of XGBoost models trained with (A) Endocrine-related cancers (N = 6: brca, ucec,prad, thca, ov, cesc)), (B) Other carcinomas (N = 9: coadread, luad, hnsc, lusc, stad, blca, lihc, kirc, kirp) and (C) Other tumors (N = 3: lgg, skcm, gbm) dataset. (D) Bubble plot of individual tumor accuracies versus corresponding tumor proportions (number of sample for each tumor type/total number of samples) within the restricted dataset (total of 18 tumor types). Colors indicate the different groups while bubble size corresponds to the individual tumor proportion within the group to which it belongs (number of sample for each tumor type/total number of samples of the corresponding biological group). (E) Charts showing the size, in terms of total count and percentage, of each group (Endocrine-related cancers, Other carcinomas and Other tumors) in the new created dataset in which each sample is associated with the group it belonged to, thus with the three groups as targets. The confusion matrix shows the performance of the model