MethylBoostER - XGBoost to classify kidney cancer subtypes

MethylBoostER is a multiclass XGBoost model that can classify pathological sub-types of renal tumours (ie, kidney cancer subtypes). This repository includes the code, the trained models and all kinds of results that demonstrate MethylBoostER's performance.

Explanation of the files

Look at the data in data/ folder:

• See UMAPs_of_data.ipynb and UMAPs/ for UMAP visualisations of the data
• See combining_EPIC_and_450k.R (and perhaps preprocess_TCGA_data.Rmd) for how we processed the training/testing set
• See preprocess_validation_data.Rmd for the preprocessing of the validation datasets
• The processed data we used are stored in this data repository: https://zenodo.org/record/6463893 (DOI 10.5281/zenodo.6463892)
  • See the README within this repo for information about this data

Train the XGBoost model:

• Run xgboost.ipynb (or xgboost.py)

See the testing set results:

• Look in figs_xgboost/.
  • Confusion matrices, ROC curves and precision recall curves are in figs/
  • Trained XGBoost models are in xgboost_models/
  • Testing set scores and metrics are in scores.json (you can read it using pickle)
  • Predictions for all samples are in all_test_predictions
• Look at xgboost_results.ipynb. This loads the saved models and re-generates the saved results figures and metrics.
  • Also generates sample number barplots for the training/testing set

Evaluate on external datasets:

• Run xgboost_external_validation.ipynb and/or look in figs_xgboost_validation/

Get high and moderate confidence predictions:

• Look at high_moderate_conf_predictions.ipynb and/or figs_xgboost_high_mod_conf/

Evaluate on multiregion samples:

• Run xgboost_multiregion_predictions.ipynb
• See figs_xgboost/figs/multiregion_predictions.svg and figs_xgboost_validation/evelonn_multiregion_predictions.svg figures

Find the features the trained XGBoost models are using:

• Look at xgboost_important_features.ipynb
• Also see map_features_to_genes.Rmd

Analyse these features:

• See feature_analysis/:
  • For gene list enrichment results, see Fishers tests.ipynb (and fdr_correction.R)
  • For visualisation of the 38 features found in all 4 models, see important_probe_plots.ipynb
  • For GREAT analysis (GO enrichment for genomic regions), see GREAT_analysis.Rmd
  • For the genomic location of the features, see Feature_analysis.Rmd (and folder gat/ for the GAT analysis results)
  • For class-specific feature analysis, see Class_specific_feature_analysis.ipynb, Class_specific_feature_analysis.Rmd and features_per_class/ (the results folder)

Look at a model trained only to classify chRCC and oncocytoma:

• See onc_vs_chRCC_experiment/ folder

Find out what happens when we vary XGBoost's max_tree_depth parameter:

• See Tree_depth_experiment.ipynb and tree_depth_experiment/ folder.

How do I find the code to generate Figure x?

• 2a: xgboost_results.ipynb

• 2b: data/UMAPs_of_data.ipynb

• 2c: xgboost_results.ipynb

• 2d: data/UMAPs_of_data.ipynb

• 2e: xgboost_results.ipynb

• 3a: high_moderate_conf_predictions.ipynb

• 3b: high_moderate_conf_predictions.ipynb

• 4a: xgboost_external_validation.ipynb

• 4b: high_moderate_conf_predictions.ipynb

• 4c,d,e and f: high_moderate_conf_predictions.ipynb and xgboost_external_validation.ipynb

• 5: xgboost_multiregion_predictions.ipynb

• 7a: feature_analysis/Feature_analysis.Rmd

• 7c: feature_analysis/GREAT_analysis.Rmd

• S1: xgboost_results.ipynb

• S2: data/UMAPs_of_data.ipynb

• S3: high_moderate_conf_predictions.ipynb

• S4a: xgboost_external_validation.ipynb

• S4b: high_moderate_conf_predictions.ipynb

• S4c: high_moderate_conf_predictions.ipynb

• S5: xgboost_external_validation.ipynb

• S6: onc_vs_chRCC_experiment/onc_vs_chRCC_hyperparameter_tuning.ipynb

• S7: data/UMAPs_of_data.ipynb

• S9: feature_analysis/important_probe_plots.ipynb

• S11a: feature_analysis/Class_specific_feature_analysis.Rmd

• S11b, c, d and e: feature_analysis/Class_specific_feature_analysis.Rmd

• S11f: feature_analysis/Class_specific_feature_analysis.ipynb

• S12a: data/UMAPs_of_data.ipynb

• S12b: data/plot_PCA_of_KIRC_expression.R

• S13: Tree_depth_experiment.ipynb

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.