Fig. 7

Schematic diagram of matrix factorization with neural network. 1) The circRNA-RBP interaction data is downloaded from the CircRic database, and the interaction matrix Y could be obtained by matching with the circRNA IDs in circBase database. 2) According to the P-U learning mechanism, negative samples with the same number of positive samples are randomly selected from the unlabeled relationships to obtain the training data set. 3) Based on the matrix classical factorization method, the neural network algorithm is used to obtain the latent factors of circRNAs and RBPs, and the scores of circRNA-RBP interactions are obtained by calculating the cosine of the latent factors. 4) The unlabeled relationships are scored using the trained model