A miRNA Target Prediction Model Based on Distributed Representation Learning and Deep Learning

Figure 4

The overall workflow. Fill the input miRNA and CTS original sequence to a uniform length, and then replace each nucleotide in the sequence with the vector trained by mi2vec and m2vec. The 50-dimension embedding of miRNA and CTS passes through the first BiLSTM layer, respectively, and then concatenates the outputs feature maps, then passes through the second BiLSTM get to 200-dimensions, and finally, uses the linear layer to reduce the dimension to 2-dimensions and prediction by softmax.