Code used in the statistical analysis for "Blood serum protein biomarkers for the screening of non-metastatic colorectal carcinomas" by Melanie Ivancic et al.
This project is set up to run in a Unix-like environment (OSX or Linux)
Using make invokes the full process of:
- Extracting the biomarker measurements from the excel files in
raw-data - Computing the per-patient-sample geometric means for each biomarker across replicates
- Matching the biomarker measurements to patient diagnoses (in
raw-data/csv/labels.csv) - Splitting the data into training and validation according to
raw-data/csv/ - Generating the pooled cross-validation ROC curves (these will be placed in a
reportsfolder) - Selecting the validation panel+method and producing ROC curves on the validation set (also in the
reportsfolder). - In the process of generating reports, pickled python objects containing the results are also generated (in a
python-resultsfolder). - In the process of generating pooled cross-validation ROC curves, statistics about how many times each method and panel is selected are printed to the command line.
Random variation: Since there is inherent randomness in the process of learning a random forest or an extremely randomized trees model, there is variation across different runs of the pipeline in the results.