Fig. 4
From: Anomaly detection in genomic catalogues using unsupervised multi-view autoencoders
Two different examples of real CRMs rebuilt in HeLa. All figures give the mean across X axis (region size) of the tensors. The model’s parameters are detailed in Additional file 10: Table S1. As with the other figures, the blue heatmap represents the original representation of the CRM, with the green heatmap giving the rebuilding by the model. The average of all HeLa CRMs is provided for comparison. The model has visibly learned different correlation groups, and not just rebuilt the average CRM. We can see, notably for BRD4, that an incomplete group results in lower scores, and that phantoms are added to complete the learned groups. Some learned groups are extracted and presented in Additional file 5: Fig. S11 for comparison