Related Experiment Video
Updated: Aug 10, 2025
04:53
Author Spotlight: Advances in Evaluating Human Lung Epithelial Cells' Response to Metal-Organic Frameworks
Published on: May 26, 2023
1.2K
Quantitative Prediction of Inorganic Nanomaterial Cellular Toxicity via Machine Learning
Nikolai Shirokii1, Yevgeniya Din1, Ilya Petrov1
1International Institute "Solution Chemistry of Advanced Materials and Technologies", ITMO University, 191002, Saint-Petersburg, Russian Federation.
Small (Weinheim an Der Bergstrasse, Germany)
|February 11, 2023
Summary
Machine learning (ML) models now quantitatively predict inorganic nanomaterial cytotoxicity, considering concentration effects. This approach enables in silico screening, reducing extensive laboratory experiments for nanomaterial development.
Area of Science:
- Materials Science
- Computational Chemistry
- Toxicology
Background:
- Machine learning (ML) has advanced organic chemistry but faces challenges in materials science, particularly in predicting biological interactions.
- Nanotoxicity assessment is crucial for bio-related research, necessitating pre-evaluation to minimize extensive experiments.
- Existing ML models offer binary toxicity predictions but lack concentration-dependent quantitative insights vital for scientists.
Purpose of the Study:
- To develop a quantitative ML model for predicting inorganic nanomaterial cytotoxicity.
- To introduce novel quantitative atom property-based descriptors for enhanced model extrapolation.
- To provide an interpretable ML model for in silico screening of nanomaterials.
Main Methods:
- Developed an ML model utilizing correlation-based feature selection and grid search for hyperparameter optimization.
- Introduced quantitative atom property-based descriptors for nanomaterial characterization.
- Employed 10-fold cross-validation (CV) to evaluate model precision (Q² = 0.86) and root mean squared error (RMSE = 12.2%).
Main Results:
- Achieved a quantitative prediction of inorganic nanomaterial cytotoxicity with high precision (Q² = 0.86, RMSE = 12.2%).
- Demonstrated the model's ability to extrapolate on unseen samples using novel descriptors.
- Identified key features through importance analysis, ensuring model interpretability.
Conclusions:
- The proposed ML approach enables accurate, quantitative prediction of nanomaterial cytotoxicity, including concentration dependencies.
- This facilitates in silico screening, significantly reducing the need for labor-intensive experimental evaluations.
- Accelerates the development of safer nanomaterials for various applications, particularly in medicine.
