Figure 6.

Improvement of performances in Rosa chinensis via pretraining and fine-tuning in the independent test. The BERT6mA models were pretrained by the whole dataset of species other than R. chinensis. The pretrained model was fine-tuned on training data of R. chinensis by 5-fold cross-validation. The fine-tuned models were evaluated on the test data of R. chinensis. The bars show the area under the curve (AUCs) of pretrained models (red), the non-pretrained model (green) and previous models (blue).