Supplementary Figure S1.

Visualization of activated features on the merge-layer extracted by DeepCAPE for training sets of different sizes (different augmentation strides). The blue and red points represent activation degree of features in DNA and DNase modules, respectively (darker color corresponds to higher activation degree). With abundant training samples (e.g., stride 1), DeepCAPE is inclined to activate only low-level features, which are extracted by the first three convolutional layers. When the sample size is limited (e.g., stride 300), however, DeepCAPE can also activate high-level features, which are extracted by the last three layers.