Figure 2
From: Identification of 12 cancer types through genome deep learning
Cancer identification performance of 12 specific models and the mixture model. (a) The classification performance of 12 specific models. Using different thresholds, the sensitivity is the abscissa and the specificity is the ordinate, resulting in 12 ROC curves. The 12 ROC curves produce perfect classification results, and the area under the ROC curve (AUC) is greater than 96%. (b) Confusion matrix of the mixture mode. The abscissa indicates the label, and the ordinate indicates the predicted cancer type. LUSC is more obvious in the predictions, especially in the BLCA predictions, suggesting that many cancers are easily confused with LUSC. Cancers that are easily confused in model predictions may be similar in their genetic variations. (c) The accuracy of top-N at different forecasted quantities. The abscissa indicates different prediction numbers, and the ordinate indicates accuracy. The accuracy of the prediction result is 70.08%, and the accuracy of two prediction results is 83.20%, which provides support for the practical application of the model. The abscissa indicates the label, and the ordinate indicates the predicted cancer type.
