Fig. 1

ReorientExpress deep learning models. ReorientExpress implements two deep neural networks (DNNs) to predict the orientation of cDNA long reads. a A multilayer perceptron (MLP) is trained on k-mer frequencies extracted from sequences of known orientation. For each test read, the orientation is predicted using the k-mer frequencies of the read as input. b A convolutional neural network (CNN) is trained on 500-nt sliding windows from sequences of known orientation, using one-hot encoding for each window. Prediction is performed by scoring all windows in a test read and calculating the mean score independently for each orientation