Skip to main content


Fig. 10 | BMC Systems Biology

Fig. 10

From: Opening up the blackbox: an interpretable deep neural network-based classifier for cell-type specific enhancer predictions

Fig. 10

t-SNE visualization of the H1 cell line for all 480 features (left) and for the top 160 features, ranked according to their importance, i.e., discriminative ability (right). In both cases, the close clustering of the positive (blue circles) and the negative (red circles) samples show that it is a difficult task to separate them. Our DNN-based architecture, through its multi-layered structure and complex tuning procedure, is able to distinguish among the different kinds of positive elements (enhancers) and the different kinds of negative elements (non-enhancers)

Back to article page