A deep learning model based on convolutional neural networks and recurrent neural networks was discovered to identify circular RNA splicing sites in the human genome by analyzing genome sequence data. First, based on the preprocessed nucleotide sequences, two Network depth, 8 convolution kernel sizes and 3 long short term memory (LSTM) parameters, a total of 16 models in 8 groups; secondly, the pooling layer is further tested for mean pooling and maximum pooling, and Adding GC content improves the prediction ability of the model; finally, the circular RNA in human seminal plasma that has been experimentally verified is predicted. The results show that the model with the convolution kernel size of 32 × 4, depth of 1, and LSTM parameter of 32 recognizes The highest rate is 0.9824 on the training set, 0.95 on the test dataset, and 83% on the experimental validation data. The model has better recognition of human circular RNA splicing sites performance.
Introduction: This is part of the unprocessed Arabidopsis data