KidneyCancerML

Machine learning models for predicting recurrence in kidney renal clear cell carcinoma (KIRC) patients using multi-omics TCGA data (mRNA, methylation, mutations, clinical).

Repository layout

data/
  raw/                          # untouched TCGA / cBioPortal exports
    kirc_tcga/                  # original cBioPortal download
    data_*.csv                  # CSV form of the raw TCGA tables
  interim/                      # preprocessing intermediates (CSVs)
  processed/                    # final arrays consumed by the models
    mrna/                       # feature matrices (.npy)
    labels/                     # recurrence labels (.npy)
    clinical/                   # age/gender/stage/grade/etc (.npy)

notebooks/
  01_preprocessing/             # raw -> interim -> processed
    txt_to_csv.ipynb              # TCGA .txt -> .csv conversion
    preprocessing.ipynb           # original mRNA preprocessing
    xscore_preprocessing.ipynb    # original multi-omics preprocessing
    zscore_preprocessing_clean.ipynb  # cleaned rewrite of the above
  02_eda/
    patient_stats.ipynb         # cohort statistics, survival, recurrence
  03_models/
    tensorflow/                 # Keras / TF models
    pytorch/                    # PyTorch models
    tpot/                       # TPOT AutoML + exported pipelines
    other/                      # Ludwig, STREAMLINE

results/
  metrics.csv / .xlsx           # cross-model comparison
  features.csv                  # selected/important features
  figures/                      # ROC curves, tables, plots

Target

Binary recurrence classification. Cleaned preprocessing uses DFS_STATUS (disease-free survival status) from the TCGA clinical file as the label; the original preprocessing.ipynb used HISTORY_OTHER_MALIGNANCY.

Data flow

1. notebooks/01_preprocessing/txt_to_csv.ipynb converts raw kirc_tcga/*.txt into the data_*.csv files under data/raw/.
2. The preprocessing notebooks read from data/raw/ and write intermediates to data/interim/ and final feature/label/clinical arrays to data/processed/.
3. Model notebooks under notebooks/03_models/ consume only the data/processed/ arrays.

Notes

• Some processed-file names contain the typo methlyation. They are kept as-is so the existing model notebooks continue to load them; the cleaned preprocessing notebook also writes to the typo'd names for the same reason.
• All notebook file paths are relative to the repo root (e.g. '../../data/raw/data_mutations.csv' from a depth-2 notebook, '../../../data/processed/mrna/processed_mrna_data_zscore.npy' from depth-3).