Skip to main content
Fig. 2 | Genome Biology

Fig. 2

From: DeepMILO: a deep learning approach to predict the impact of non-coding sequence variants on 3D chromatin structure

Fig. 2

Datasets of anchors and non-anchors, and performance of models on these datasets. Numbers inside brackets in c and d are AUPRCs. a Different types of non-anchors used for training and testing. b The CNN model outperformed the RNN model, and their combination delivered the best performance for the non-anchor type 1 test set. Proportion of positive samples is 0.45. c Architecture of the anchor model consisting of a CNN and an RNN. d Performance of the anchor model (CNN + RNN) on test sets. Proportions of positive samples in datasets are 0.45, 0.12, and 0.075 for non-anchor types 1, 2, and 3, respectively

Back to article page