Targeted Metabolomics based Brain Tumor Margin Assessment

Source code for "Targeted Metabolomics Analyses for Brain Tumor Margin Assessment During Surgery"

Dependencies

The conda environment is given as hrmas.yml if you wish to clone the environment to your machine. Please consider using our environment to avoid any inconvenience resulted from version differences.

• PyTorch
• scikit-learn
• pandas
• numpy
• xlrd
• PyNMR
• shap

Getting Started

Predict with your data

• Go to /predict_with_your_data/config.py and change path variable and write the path to that folder and change the lib variable to the path to pyNMR library.
• Please run the code with the anonymous sample and see the prediction printed out.
• Move your dataset folder to /predict_with_your_data/data folder.
• If your data is aligned to lactate and normalized to acetate then please keep shift_ and normalization_ variables untouched. Else, if the alignment and normalization of your data is different than that, please change shift_ and normalization_ variables accordingly. We recommend -1515 and 1.6 respectively, if your data is aligned to water.
• You can run /predict_with_your_data/predict.py, and see the predictions printed out to your console.
• The script will predict whether the samples in your dataset are of control, benign glioma or aggressive glioma tissue. You will see the predictions with their corresponding probability.

There are 3 main trainings you may do in this repository. These are automated metabolite quantification, pathologic classification and sample visualization.

Train with our dataset

• Download the dataset from here. Extract the compressed folder and move all the contents of cpmg_dataset folder to /train_with_your_data/data/cpmg folder and move all the contents of eretic_dataset folder /train_with_your_data/data/eretic_cpmg as a subdirectory.
• Download the Supplementery Table 2 and rename it to Table_S2.xls, then move it into data_xlsx folder.
• Go to scripts/config_ult.py and change the base variable and write the path to that folder and change the lib variable to the path to pyNMR library.
• Run create_processed_cpmg_dataset.py and create_processed_eretic_cpmg_dataset.py scripts.
• pathologic classification training will require models of automated metabolite quantification. Go to the automated metabolite quantification folder and train all the models for all metabolites by running train.py script.
• Go to pathologic classification and train both the benign vs aggressive and control vs tumor classification models by train.py script under the model of interest. We recommend you to use the /cpmg/pathologic_classification/control_tumor/metabolites/predicted/all_metabolites/RF and /cpmg/pathologic_classification/benign_aggressive/metabolites/predicted/all_metabolites/RF.

Train with your data

• If your dataset consists of CPMG samples, move the contents of that dataset into /train_with_your_data/data/cpmg folder, likewise if your data consists of ERETIC-CPMG samples, move the contents of that dataset into /train_with_your_data/data/eretic_cpmg folder.
• The model training requires an xls file, and we release our file in Supplementery Table 2. Please format your spreadsheet according to our Table.
• Go to scripts/config_ult.py and change the base variable and write the path to that folder and change the lib variable to the path to pyNMR library.
• Run create_processed_cpmg_dataset.py and create_processed_eretic_cpmg_dataset.py scripts.
• Apply the steps 4 and 5 of Train with our dataset.