BoBaFor

• a bacterial genome wide association approach through utilizing machine learning practices.

installation

Create Conda Virtual Environment with required python version

conda create -n BoBaFor_env python==3.11.10

Activate BoBaFor_env

conda activate BoBaFor_env

install BoBaFor via pip

pip install BoBaFor

Use

Download example from GitHub

git clone https://github.com/PaulDanPhillips/BoBaFor.git

change directory to example

cd BoBaFor/examples

Convert relative paths to absolute in example_config.yaml

python AddAbsolutePaths.py

Run the example data (user needs to pay attention to how many cores they have available and want to use)

BoBaFor --config example_config.yaml --cores 4

Running your own data:

Data organization

• The data needs to be split into two files: 1 containing all of the genetic features and 2 the response phenotype ensureing that the indexes on the feature matrix align with the reponse vector.

Configure.yaml file:

• You will edit the Sim1.yaml file from the example directory.

1. Predictor
   • The absolute or relative path to your genetic feature matrix will take the place of """balanced_features/Strongest_Sim1.txt"""
   • /path/to/data/features.txt
2. Response
   • The absolute or relative path to your response phenotype vector will take the place of """balanced_responses/Sim1.txt""" /path/to/daata/response.txt
3. *chi2Correct
   • The user need to decie if they want to the chi2 prefilter step to be corrected via false-discovery-rate or not (only True or False)
   • True/False
4. FDRchi2Thresh
   • The user needs to decide what the FDR chi2 threshold is for removing genetic features based on this chi2 test is.
   • 0.-1. (float)
5. PreFilterFileName
   • The name you want for the output files from just the chi2-prefilter step.
   • Prefilter_Experiment_Name
6. sim
   • Experiment Name for prefix file name
   • Simulation1
7. GridParamDepth
   • The number of of options to keep for EVERY hyperparameter after optimizing from RandomSearch leading into the much more thorough and computational demanding GridSearch
   • 1-5 (int)
8. ScoreMetric
   • What metric to use to score the model. Pay attention to what type of data your response metric is and how balanced the data is.
   • balanced_accuracy
9. RandSearch1_Niter
   • The number of RandomSearch iterations to perform prior to boruta selection (The number of hyperparamter combinations to test)
   • (int)
10. GridSearch1_CV
    • The number of GridSearch iterations to perform prior to boruta selection (The number of times to create kfold splits and test EVERY possible hyperparameter combination)
    • (int)
11. boruta_perc
    • The percent strength of the boruta shadow-features to be comparted against the real features.
    • 0-100 (int)
12. boruta_pval
    • Level at which the corrected p-values will get rejected in both correction steps.
    • 0.-1. (float)
13. borutaSelect_perc
    • The percent theshold of which to select final features (the number of times a feature was selected divided by the number of iterations boruta was run).
14. collect
    • A boolean on whether to keep all selected features no matter the boruta_Select_perc setting.
    • True/False
15. boruta_reps
    • The number of iterations to run boruta selection.
16. model
    • Whether to run a random forest or extreme gradient boosted model
    • RF/XGB
17. RandSearch2_Niter
    • The number of RandomSearch iterations to perform after boruta selection (The number of hyperparamter combinations to test)
    • (int)
18. RandSearch2_CV
    • TO BE REMOVED
    • The number of iterations to create kfolds within a single RandSearch2_Niter.
    • (int)
19. GridSearch2_CV
    • The number of GridSearch iterations to perform after boruta selection (The number of times to create kfold splits and test EVERY possible hyperparameter combination) (int)
20. RankingFeatureIters
    • The number of iterations to perform feature ranking via Feature Importance or Permutation Importance
    • (int)
21. PermImpInteralIter
    • The number of internal repetitions for Permutation Importance
    • (int)

Output files:

• Most important is BorutaOUT_1.00.05_RF/FinalFeatures_RF/eatureImportance_Ranked_*.out as it contains the genentic variants and their associated feature importance value.

  1. BorutaOUT_1.00.05_RF/PreFilterOut/PreFilterPredictor_Sim1.txt
     • The features that pass the chi2 prefilter set in the config.yaml file
  2. BorutaOUT_1.00.05_RF/TuningOUT_RF/GridParamSpace1_Sim1.pickle
  3. BorutaOUT_1.00.05_RF/TuningOUT_RF/GridParamSpace2_Sim1.pickle
  4. BorutaOUT_1.00.05_RF/TuningOUT_RF/gridSearch1_Sim1.txt
  5. BorutaOUT_1.00.05_RF/TuningOUT_RF/gridSearch2_Sim1.txt
  6. BorutaOUT_1.00.05_RF/TuningOUT_RF/GridSearchFIG1_Sim1.png
  7. BorutaOUT_1.00.05_RF/TuningOUT_RF/GridSearchFIG2_Sim1.png
  8. BorutaOUT_1.00.05_RF/TuningOUT_RF/Optimal_Tuning1_Sim1.pickle
  9. BorutaOUT_1.00.05_RF/TuningOUT_RF/Optimal_Tuning2_Sim1.pickle
     • The optimized model used to rank the genetic features. This can be used with the explainable AI shap to best interpret the model if desired.
  10. BorutaOUT_1.00.05_RF/TuningOUT_RF/randSearch1_Sim1.txt
  11. BorutaOUT_1.00.05_RF/TuningOUT_RF/randSearch2_Sim1.txt
  12. BorutaOUT_1.00.05_RF/BorutaOUT/RF_Boruta_Selected_Sim1.txt
  13. BorutaOUT_1.00.05_RF/BorutaOUT/RF_Boruta_Tentative_Sim1.txt
  14. BorutaOUT_1.00.05_RF/BorutaOUT/RF_FinalFeatureMatrix_BorutaSelected_Sim1.txt
  15. BorutaOUT_1.00.05_RF/FinalFeatures_RF/FeatureImportance_Ranked_Sim1.out