Skip to main content
Fig. 4 | BMC Bioinformatics

Fig. 4

From: Unsupervised deep learning reveals prognostically relevant subtypes of glioblastoma

Fig. 4

Consensus clustering of all samples. (a) Consensus clustering of the 3rd hidden layer representations from our DBN model (model 5 in Table 2) captured tissue-specific clustering. The cancer types within each cluster are shown on top. (b) Composition of each of the clusters learned in part a, including the number of samples of each cancer type in each cluster. (c) Consensus clustering results when simply using the high-dimensional (9476 features/genes) raw data without any dimensionality reduction. (d) Consensus clustering results for high-dimensional (7160 features/genes) raw data without any dimensionality reduction. All heatmaps evaluated at k = 14 (number of clusters). Dark blue corresponds to samples that always cluster together (consensus value = 1). White corresponds to samples that never cluster together (consensus value = 0). Lighter shades of blue are intermediate between 1 and 0

Back to article page