Volume 17, Issue 4 (Winter 2026)
2026, 17(4): 87-98 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Rahdar A, Shirzad M, Pourmadadi M. Predicting Key Properties of 5-Fluorouracil Anticancer Drug Carrier Nanocomposites Using Machine Learning: A Multi-Objective Approach. North Khorasan University of Medical Sciences 2026; 17 (4) :87-98
URL: http://journal.nkums.ac.ir/article-1-3321-en.html
URL: http://journal.nkums.ac.ir/article-1-3321-en.html
1- Department of Physics, University of Zabol, Zabol, Iran , a.rahdar@uoz.ac.ir
2- Department of Medical Nanotechnology, School of Advanced Technologies in Medicine, Tehran University of Medical Sciences, Tehran, Iran
3- Protein Research Center, Shahid Beheshti University, Tehran, Iran
2- Department of Medical Nanotechnology, School of Advanced Technologies in Medicine, Tehran University of Medical Sciences, Tehran, Iran
3- Protein Research Center, Shahid Beheshti University, Tehran, Iran
Abstract: (19 Views)
Introduction: The drug 5-fluorouracil is commonly used in cancer treatment, but its clinical effectiveness is limited due to systemic toxicity, a short half-life, and insufficient tumor tissue uptake. The design of 5-fluorouracil carrier nanocomposites should focus on maximizing loading efficiency (LE%) and encapsulation efficiency (EE%) while minimizing cytotoxicity to normal cells. The aim of this study was to develop a multi-objective machine learning model to simultaneously predict LE%, EE%, and normal cell toxicity values in 5-fluorouracil nanocarriers.
Methods A dataset comprising 20 types of nanocomposites was compiled from reputable scientific sources. Six regression models and four classification models were trained using cross-validation. Model interpretability was achieved through SHAP analysis. Virtual screening was then performed on 120 hypothetical formulations to identify the optimal candidates. The XGBoost model exhibited the best regression performance for predicting LE% (R²=0.91, MAE=1.82) and EE% (R²=0.89, MAE=2.05), while the Random Forest algorithm outperformed others in cytotoxicity classification (F1=0.88, AUC=0.92).
Results: SHAP analysis revealed that chitosan, polyethylene glycol (PEG), and graphitic carbon nitride (g-C₃N₄) enhance encapsulation efficiency, whereas halloysite nanotubes, particle sizes larger than 400 nm, and high positive zeta potential increase toxicity. The optimal design range was determined to be 100-300 nm particle size and a zeta potential of +20 to +40 mV. The formulation containing chitosan/PEG/g-C₃N₄/hydroxyapatite was identified as the top candidate, with a predicted EE% of approximately 89.7% and low toxicity.
Conclusion: The machine learning-based system provides valuable predictive and mechanistic insights for designing safe and efficient 5-fluorouracil nanocarriers. Despite data limitations, this approach shows the strong potential of computational methods in nanodrug development
Methods A dataset comprising 20 types of nanocomposites was compiled from reputable scientific sources. Six regression models and four classification models were trained using cross-validation. Model interpretability was achieved through SHAP analysis. Virtual screening was then performed on 120 hypothetical formulations to identify the optimal candidates. The XGBoost model exhibited the best regression performance for predicting LE% (R²=0.91, MAE=1.82) and EE% (R²=0.89, MAE=2.05), while the Random Forest algorithm outperformed others in cytotoxicity classification (F1=0.88, AUC=0.92).
Results: SHAP analysis revealed that chitosan, polyethylene glycol (PEG), and graphitic carbon nitride (g-C₃N₄) enhance encapsulation efficiency, whereas halloysite nanotubes, particle sizes larger than 400 nm, and high positive zeta potential increase toxicity. The optimal design range was determined to be 100-300 nm particle size and a zeta potential of +20 to +40 mV. The formulation containing chitosan/PEG/g-C₃N₄/hydroxyapatite was identified as the top candidate, with a predicted EE% of approximately 89.7% and low toxicity.
Conclusion: The machine learning-based system provides valuable predictive and mechanistic insights for designing safe and efficient 5-fluorouracil nanocarriers. Despite data limitations, this approach shows the strong potential of computational methods in nanodrug development
Keywords: 5-fluorouracil, Cytotoxicity Prediction, Encapsulation Efficiency, Multi-Objective Machine Learning, Random Forest, SHAP Analysis, XGBoost
Type of Study: Orginal Research |
Subject:
Basic Sciences
Received: 2025/11/2 | Accepted: 2025/12/22 | Published: 2026/01/1
Received: 2025/11/2 | Accepted: 2025/12/22 | Published: 2026/01/1
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |