Related Experiment Video
Updated: Oct 9, 2025
02:52
Small Molecule Screening and Toxicity Testing in Early-stage Zebrafish Larvae
Published on: March 7, 2025
1.5K
Identifying diverse metal oxide nanomaterials with lethal effects on embryonic zebrafish using machine learning
Richard Liam Marchese Robinson1, Haralambos Sarimveis2, Philip Doganis2
1School of Chemical and Process Engineering, University of Leeds, Leeds, LS2 9JT, United Kingdom.
Beilstein Journal of Nanotechnology
|December 22, 2021
Summary
Machine learning models predict nanomaterial toxicity in zebrafish embryos. A simple descriptor, Pauling electronegativity, effectively predicted lethality, simplifying safety assessments for "safe by design" approaches.
Area of Science:
- Nanotoxicology
- Computational Toxicology
- In vivo Ecotoxicology
Background:
- Manufacturers require predictive models for nanomaterial safety to support "safe by design" strategies.
- Embryonic zebrafish (Danio Rerio) are increasingly utilized as a relevant in vivo model for human safety assessment.
- Existing methods often face late-stage attrition due to insufficient predictive safety data.
Purpose of the Study:
- To develop machine learning models for predicting metal oxide nanomaterial lethality in embryonic zebrafish.
- To identify key descriptors for predicting toxicity endpoints at low concentrations (≤250 ppm).
- To evaluate the predictive performance of models focusing on excess lethality at 120 hours post-fertilisation.
Main Methods:
- Utilized data from the Nanomaterial Biological-Interactions Knowledgebase for 44 metal/metalloid oxide nanomaterials.
- Developed and evaluated machine learning models using nested cross-validation.
- Compared models using multiple descriptors (core, shell, surface, particle characteristics) against single descriptors, including Pauling electronegativity.
Main Results:
- Models predicting lethality at 24 hours post-fertilisation were less accurate, potentially due to varied exposure routes.
- A single descriptor, the Pauling electronegativity of the metal atom, achieved predictive performance comparable to models using multiple complex descriptors for 120-hour excess lethality.
- Data augmentation techniques did not improve predictive performance in this nano-Quantitative Structure-Activity Relationship (nano-QSAR) study.
Conclusions:
- Pauling electronegativity is a surprisingly effective descriptor for predicting metal oxide nanomaterial lethality in zebrafish embryos.
- This finding simplifies nano-QSAR modeling for certain toxicity endpoints, supporting the "safe by design" paradigm.
- Further research is needed to validate these findings with external data and explore more sophisticated descriptors.
