Development of a Consensus-Based QSAR Models for Enhanced Multi-Endpoint Prediction of Systemic and Local Toxicities

FAUZAN SYARIF NURSYAFI¹, MUHAMMAD ADNAN PRAMUDITO², YUNENDAH NUR FUADAH³, and KI MOO LIM¹,⁴,⁵**
¹Computational Medicine Lab, Department of Medical IT Convergence Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea
²Computational Medicine Lab, Department of IT Convergence Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea
³Telecommunication Engineering Study Program, School of Electrical Engineering, Telkom University Main Campus, Bandung, Indonesia
⁴Computational Medicine Lab, Department of Biomedical Engineering, Kumoh National Institute of Technology, Gumi, 39177, Republic of Korea
⁵Meta Heart Co., Ltd, Gumi, 39253, Republic of Korea

Corresponding authors: kmlim@kumoh.ac.kr, yunendah@telkomuniversity.ac.id

---

Overview

This repository contains the code for developing and evaluating QSAR (Quantitative Structure–Activity Relationship) models for multi-endpoint chemical toxicity prediction. The models employ a consensus-based approach integrating:

• Multiple molecular fingerprints: Morgan, MACCS, APF
• Physicochemical descriptors
• Machine learning algorithms: Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM)

Cross-validation ensures robust and reliable model performance.

---

Features

• Data preprocessing and descriptor calculation
• Model training using both fingerprint and physicochemical features
• Cross-validation with manual and original splits
• Evaluation of both individual and consensus models
• Datasets included for training and evaluation:
  • Manual split dataset
  • Original split dataset
• Ensures reproducibility with fixed package versions

---

Dependencies

• Python 3.x
• RDKit (v2025.3.2)
• CDK-pywrapper (v0.1.1)
• TensorFlow (v2.19.0)
• scikit-learn (v1.6.1)
• NumPy (v2.1.3)
• Pandas (v2.2.3)
• Additional: install-jdk (v0.3.0), bounded-pool-executor (v0.0.3)

---

Usage

1. Compute Descriptors

Use Descriptor Computation_Preprosesing data.ipynb to calculate molecular fingerprints and physicochemical descriptors for your dataset.

2. Train Models

• Training_ML_Fingerprint_10foldCrossvalidation.ipynb for fingerprint-based models
• Training_ML_PhysicochemicalProperties_10foldCrossvalidation.ipynb for physicochemical descriptor-based models

3. Evaluate Models

Use Model_Evaluation.ipynb to perform consensus evaluation and generate metrics for all models.

---

Citation

If you use this repository in your research, please cite:

Development of a Consensus-Based QSAR Models for Enhanced Multi-Endpoint Prediction of Systemic and Local Toxicities
FAUZAN SYARIF NURSYAFI¹, MUHAMMAD ADNAN PRAMUDITO², YUNENDAH NUR FUADAH³*, and KI MOO LIM¹,⁴,⁵*