4.4.4 Saliency - WangLabTHU/GPro GitHub Wiki
hcwang and qxdu edited on Aug 4, 2023, 1 version
Saliency Map is an important concept of deep learning and Computer vision. While training sequences of DNA, how does predictor model knows to focus on high-expresssion position and ignore the flanking sequences and the other background noise in the sequences? By using the concept of Saliency Map.
Caution: Note that the format of the sequence and expression files here should be consistent with the QuickStart section.
| params | description | default value |
|---|---|---|
| predictor | the trained predictor class | |
| predictor_modelpath | the pretrained model checkpoint, should be "x/xxx.pth" format | |
| predictor_training_datapath | path of natural sequences, training set for predictor will be the best | |
| predictor_expression_datapath | path of corresponding expression level with predictor_seqpath
|
|
| report_path | saving folder | |
| file_tag | saving name | |
| num_seqs_to_test | sampling scales for frequency comparison | 200 |
from gpro.evaluator.saliency import plot_saliency_map
project_path = "your project path"
predictor_training_datapath = os.path.join(project_path,'data/diffusion_promoter/sequence_data.txt')
from gpro.predictor.densenet.densenet import DenseNet_language
predictor = DenseNet_language(length=50)
predictor_modelpath = os.path.join(project_path, 'checkpoints/densenet/' + 'checkpoint.pth')
plot_saliency_map(predictor, predictor_training_datapath, predictor_modelpath,
report_path="./results/", file_tag="DenseNet")The final result will be saved in the ./results directory.