Figure 3.

Statistic results of iDRO. (A) Accuracy of different BiLSTM-CRF hyperparameter setup. Different groups represent different hidden layer dimension and colors represent different learning rate. The setup with 200 dimensions and 0.003 learning rate has the highest accuracy (B) Machine translation indicators for different preprocessing method. The combination of Codon Box and Special Tokens has the highest score for both Jaccard and Rouge. Jaccard and Rouge are commonly used to indicate the similarity between two sentences (C) Machine translation indicators of different learning rate. In both 5′UTR generation and 3′UTR generation, RNA-Bart with learning rate has the best performance.