The convolutional neural network (CNN), trained previously on 160,000 regulatory sequences obtained from glioma samples from 36 individuals, was considered and retrained for 100 epochs using sequences of elements identified in human iAstrocytes. Training group sequences were classified as: promoter active, promoter inactive, non-promoter active, and non-promoter inactive (n= 40,000 sequences in each class). The re-trained network's area under the curve (AUC) was increased after re-training.
