RPCA_hotspot

Description:

Codes for the paper RBHS: https://doi.org/10.1002/prot.26047, a robust principal component analysis based method to predict preotein-protein interaction hot spots.

Background:

Proteins often exert their function by binding to other cellular partners. The hot spots are key residues for protein-protein binding. Their identification may shed light on the impact of disease associated mutations on protein complexes and help design protein-protein interaction inhibitors for therapy. Unfortunately, current machine learning methods to predict hot spots, suffer from limitations caused by gross errors in the data matrices. This has the codes for a novel data pre-processing pipeline that overcomes this problem by recovering a low rank matrix with reduced noise using Robust Principal Component Analysis. It further trains and validates machine learning classifiers to existing databases to show the predictive power of the method RBHS.

Data provided:

• Training Data is at:
  "train_HB34.xls"
  train_HB34.xls is the file used for training the classifiers. It contains 313 rows indicating 313 interface residues and 59 columns, where 1-58 columns are features and column 59 is the label values.
• Test Data is at:
  "test_BID18.xlsx"
  test_BID18.xlsx is the independent test set file. It contains 126 rows indicating 126 interface residues and 59 columns, where 1-58 columns are features and column 59 is the label values.

Environment file is at:

"HotSpots.yml"
: All the packages used for RBHS.

File organisation:

Generating baseline results using original data and PCA

• PCA Baseline results is at :
  "PCA_baseline.ipynb"
  : This jupyter notebook can be used to generate the PCA baseline results that are used to compare the efficacy of RBHS with PCA in the paper https://doi.org/10.1002/prot.26047.

• Random Forests Classifier Baseline results are at :
  "RandomForestClassifier_Baseline.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning Random forest classifier on HB-34.xls dataset without any preprocessing. It can then be used for testing the tuned random forest classifier on BID-18.xlsx without any preprocessing done on it.

• Support Vector Machine Baseline results are at :
  "gridsearchSVM_Baseline.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning SVM classifier on HB-34.xls dataset without any preprocessing. It can then be used for testing the hyperparameter-tuned SVM classifier on BID-18.xlsx without any preprocessing done on it.

• Gradient Boosting Machine Baseline results are at :
  "tuningGBM_Baseline.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning GBM classifier on HB-34.xls dataset without any preprocessing. It can then be used for testing the trained GBM classifier on BID-18.xlsx without any preprocessing done on it.

• Extreme Gradient Boosting Baseline results are at :
  "tuningXGB_B.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning GBM classifier on HB-34.xls dataset without any preprocessing. It can then be used for testing the trained GBM classifier on BID-18.xlsx without any preprocessing done on it.

RBHS: Robust principal component analysis-based prediction of protein-protein interaction hot spots

• Codes for RBHS pipeline are at :
  "reading_hotspot2Data_v4.m"
  : This matlab file is used to generate data for all the classifiers implemented in the jupyter notebooks.
• Codes for RPCA are at :
  "robustPCA_v3.m"
  : This matlab file contains the function to implement robust principal component Analysis (RPCA).

Results for RBHS+classifiers

• Random Forests Classifier results are at :
  "RandomForestClassifier.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning Random forest classifier on HB-34.xls dataset with RBHS applied to it. It can then be used for testing the tuned random forest classifier on BID-18.xlsx with RBHS applied to it.

• Support Vector Machine results are at :
  "gridsearchSVM.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning SVM classifier on HB-34.xls dataset with RBHS applied to it. It can then be used for testing the hyperparameter-tuned SVM classifier on BID-18.xlsx with RBHS applied to it.

• Gradient Boosting Machine results are at :
  "tuningGBM.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning GBM classifier on HB-34.xls dataset with RBHS applied to it. It can then be used for testing the trained GBM classifier on BID-18.xlsx with RBHS applied to it.

• Extreme Gradient Boosting results are at :
  "tuningXGB.ipynb"
  : This jupyter notebook can be used for training and hyperparameter tuning XGBoost classifier on HB-34.xls dataset with RBHS applied to it. It can then be used for testing the trained XGBoost classifier on BID-18.xlsx with RBHS applied to it.

Generating Plots

• Plotting of precision-recall curves for all algorithms can be done using :
  "plot_curves_prec_rec.ipynb"

• Plotting of ROC curves for all algorithms can be done using :
  "plot_curves_roc.ipynb"

Citation

If you find this repo useful and would like to cite it, please use the following citation:

@software{Sitani_RBHS_Sitani_2023,
author = {Sitani, Divya},
month = aug,
title = {{RBHS_Sitani}},
url = {https://github.com/Divya1205/RBHS_Sitani},
version = {1.0.0},
year = {2023}
}

If you would like to read and cite the publication RBHS: https://doi.org/10.1002/prot.26047, a robust principal component analysis based method to predict preotein-protein interaction hot spots:

@article{sitani2021robust,
  title={Robust principal component analysis-based prediction of protein-protein interaction hot spots},
  author={Sitani, Divya and Giorgetti, Alejandro and Alfonso-Prieto, Mercedes and Carloni, Paolo},
  journal={Proteins: Structure, Function, and Bioinformatics},
  volume={89},
  number={6},
  pages={639--647},
  year={2021},
  publisher={Wiley Online Library}
}