ganjf / biomarkerprediction Goto Github PK

Python Environment: pip install -r requirements.txt

Install MMDetection following https://github.com/open-mmlab/mmdetection.

Install SLS code: git clone https://github.com/ganjf/biomarkerPrediction.git

Set the biomarker documentation with format following TMB_CPTAC_LUAD.txt.

The WSI demo case could be downloaded from TCGA-44-8117-01A-01-BS1.

Modify the source code.

• ./subtypeDetection/inference/processSlideWindow.py
  • line 99-112: Set the contents of WSIs which need to be processed.
• ./subtypeDetection/inference/makeDataset.py
  • line 45-46 and line 51-52: Set the mapping releationship between WSI name and patient id presented in Biomarker documentation.

Modify the executable of subtypeTest.sh.

• --wsi_dir: The content of WSIs in test set.

• --patch_dir: The content to store the patches (for detection) generated by fixed-size sliding window method.

• --biomarker_txt: The path of biomarker documentation.

• --inference_coco: The test input for detector in COCO format, summarizing the ensembles of $patch_dir.

  In cascade_rcnn.py line 268: set 'ann_file' = $inference_coco

  line 269: set 'imp_predix' = $patch_dir

• --detection_ckp: The checkpoint for tumor search model.

• --result_pkl: The .pkl path to store detection results.

• --roi_json: The .json file to store the result after mapping detetcion results to the raw WSI.

• --roi_dir: The content to store the post-processed tumor tissues.

• --data_test: The .csv file to summarize the ensembles of $roi_dir.

Run bash subtypeTest.sh

Modify the source code.

• ./TMB/ood_evaluate.py
  • line 50-51: Set the mapping releationship between WSI name and patient id presented in Biomarker documentation.
• ./TMB/ood_screen.py
  • line 9-10, line 14-15, line 21-22 and line 36-37: Set the mapping releationship between WSI name and patient id presented in Biomarker documentation.
• ./TMB/cls_evaluate.py
  • line 49-50: Set the mapping releationship between WSI name and patient id presented in Biomarker documentation.

Modify the executable of biomarkerTest.sh

• --data_dir: The content of the post-processed tumor tissues (as $roi_dir).
• --biomarker_txt: The path of biomarker documentation.
• --input_test: The .csv file to summarize the post-processed tumor tissue associated to the patients in test set.
• --confidence_ckp: The checkpoint for ROB search model.
• --biomarker: The choices of ['TMB', 'CD274', 'CD8A', 'TP53']
• --confidence_threshold: The threshold to perform ROB searching, default=55(%).
• --biomarker_threhold: The threshold to determine the status of biomarker as high/positive.
• --confidence_input_test: The .csv file to store the biomarker predictions and confidence scores from ROB search module.
• --confidence_output_test: The .csv file to summarize the screened-out ROBs from ROB search module.
• --cls_ckp: The checkpoint for status prediction model.

Run bash biomarkerTest.sh >> test.log