Related Experiment Video
Updated: Aug 13, 2026
09:01
A High-throughput Assay for the Prediction of Chemical Toxicity by Automated Phenotypic Profiling of Caenorhabditis elegans
Published on: March 14, 2019
A Multimodal Generative AI Framework for Predicting the Toxicity of Nanoparticles
Leonid Legashev1, Arthur Zhigalov1, Irina Bolodurina1
1Research Institute of Digital Intelligent Technologies, Orenburg State University Named After V.A. Bondarenko, Pobedy Pr. 13, Orenburg 460018, Russia.
Nanomaterials (Basel, Switzerland)
|August 12, 2026
Summary
This study introduces a generative AI framework to design safer nanoparticles by predicting cytotoxicity. The best model, TabDDPM, accurately predicts nanoparticle toxicity and guides the development of biocompatible nanomaterials.
Area of Science:
- Nanotechnology
- Computational Chemistry
- Artificial Intelligence
Background:
- Predicting engineered nanoparticle (ENP) cytotoxicity is difficult due to diverse properties.
- Developing ENPs with controlled toxicity is crucial for safe applications.
Purpose of the Study:
- To create a multimodal generative framework for synthesizing ENP candidates with specific toxicity.
- To benchmark generative architectures for capturing structure-activity relationships in ENP toxicity.
Main Methods:
- Utilized large language models and SciBERT embeddings to extract and encode toxicity data from literature.
- Benchmarked four generative architectures (CTGAN, TVAE, WGAN-GP, TabDDPM) using Synthetic Data Vault (SDV) quality score.
- Validated the top model (TabDDPM) using coarse-grained molecular dynamics simulations (GROMACS, Martini force field).
Main Results:
- TabDDPM achieved the highest SDV quality score (0.78), demonstrating superior performance in capturing structure-activity relationships.
- Simulations revealed toxic ENPs induce higher electrostatic stress and prolonged membrane equilibration.
- Safe ENPs exhibited reduced nanoparticle-membrane hydrophobic core interaction compared to toxic ones.
Conclusions:
- The generative framework accurately replicates statistical distributions and captures biophysical mechanisms of membrane disruption.
- This approach offers a robust tool for in silico design of biocompatible nanomaterials.
- The study advances the prediction and design of safer nanoparticles.
