Remission Classifier: A Machine Learning Approach for Clinical Predictions

This repository contains the code and environment configuration used in the paper titled: PRECISE-RA: Predicting Remission and Stratifying Risk in Rheumatoid Arthritis Patients Treated with bDMARDs—A Robust Machine Learning Approach

Overview

The repository includes preprocessing steps, model training, validation, and calibration analysis as described in the paper.

Files

• RemissionClassifier.ipynb: Main notebook for training machine learning models, hyperparameter tuning, cross-validation, evaluation, calibration, and risk stratification.
• preprocess.ipynb: Notebook for preprocessing the clinical data, including feature engineering, missing value handling, and data alignment.
• environment.yml: Conda environment file to recreate the computational environment used for this project.

Features

• Data Preprocessing:

  • Handles clinical data with missing values and prepares it for analysis.
  • Performs feature engineering and alignment for training and testing datasets.

• Model Training and Evaluation:

  • Implements machine learning models, including XGBoost, AdaBoost, SVM and Random Forest.
  • Hyperparameter tuning using GridSearchCV.

• Calibration Methods:

  • Includes Platt scaling (Sigmoid), Isotonic regression, Beta calibration, and Spline calibration.

• Performance Metrics:

  • Accuracy, precision, recall, F1-score, Matthews correlation coefficient (MCC), Brier score, and ROC-AUC.

• Visualization Tools:

  • Generates calibration curves, ROC curves, and confusion matrices for internal and external validation.

How to Use

1. Clone the repository:

   git clone https://github.com/fatemehsalehi/PRECISE-RA.git

Set up the environment:

Using Conda (recommended):

conda env create -f environment.yml conda activate RAproject Using Virtual Environment:

python -m venv RAproject source RAproject/bin/activate # On Windows: RAproject\Scripts\activate pip install -r requirements.txt

Open and run the notebooks using Jupyter Notebook or JupyterLab:

jupyter notebook Replace dataset.csv and dataset_test.csv with your datasets (structured as described in the paper).

Dependencies The dependencies will be automatically installed when setting up the environment.

Dataset The repository does not include the datasets used in the study due to privacy concerns. The test dataset can be found on Zenodo with the identifier doi:10.5281/zenodo.12507169. Ensure your datasets are structured as described in the paper when running the code.

Contact For questions or further information, please contact:

Fatemeh Salehi (author) - [email protected]