Fig. 7

The tenfold cross-validation accuracy of a binary SVM with a linear kernel trained on the 100-dimensional latent space for each body site (A) and methylation status (B) for gastric adenocarcinoma. The linear kernel is used to test the separability of each in the full 100-dimensional latent space. The high performance of these models indicates that the body sites and methylation statuses are not overlapping in the higher dimensional latent space even though it may appear so in lower dimension representations (Fig. 6)