This project was developed as part of the Microsoft AI National Skilling Initiative (AINSI) Internship and focuses on predicting Adverse Drug Reactions (ADRs) from chemical compound data using machine learning.
Adverse Drug Reactions (ADRs) are one of the key challenges in drug safety. This project applies machine learning to predict potential ADRs using a dataset containing drug compounds represented in SMILES format.
- Source: Kaggle
- File:
adr_dataset.csv - Content: Contains chemical compounds (as SMILES) and multiple associated ADRs.
- Purpose: To explore patterns between chemical structure and side effects for drug safety enhancement.
- Python
- Google Colab
- RDKit – For chemical fingerprint generation
- Scikit-learn – For machine learning models
- Pandas & NumPy – Data handling
- Matplotlib/Seaborn – Visualization
- Algorithm: Random Forest wrapped with MultiOutputClassifier
- Features: Molecular fingerprints derived from SMILES
- Labels: Multi-label ADR categories
- Goal: Predict which ADRs are likely for a given drug compound
The model showed high performance in terms of:
- Accuracy
- F1 Score (micro)
- Hamming Loss
Visual comparisons between predicted and actual ADRs supported the model’s effectiveness.
This project proved the viability of using chemical fingerprints for ADR prediction. It supports safer drug development and strengthens pharmacovigilance systems.
- Use larger, more diverse datasets
- Integrate clinical, demographic, and dosage data
- Apply deep learning (e.g., GNNs or Transformers)
- Real-time deployment using cloud or edge platforms
adr_dataset.csv– Dataset fileadr_prediction.ipynb– Google Colab notebookADR_Prediction_Report.txt– Model summary reportREADME.md– Project overviewrequirements.txt– Required Python packages
- Devarshi Hatwar
This project is created under the Microsoft AINSI Internship and is intended for academic and research purposes only.