Using explainable AI to predict how iron-based nanoparticles affect cell health

What is it about?

Iron carbide nanoparticles are promising materials for use in medicine, for example in drug delivery, imaging, and cancer treatment. However, understanding whether they are safe for cells is essential before they can be widely used. In this study, we developed an artificial intelligence (AI) approach to predict how these nanoparticles affect cell survival (cytotoxicity). Instead of relying only on experimental measurements, we used detailed “atom-level descriptors,” which capture how the nanoparticles are structured at the atomic scale. These descriptors were combined with machine learning models to improve prediction accuracy. Two different modeling strategies were tested, and the atom-based approach showed better performance when applied to enriched data. The final model, based on a Random Forest algorithm, follows international (OECD) principles for reliable predictive models.  Importantly, we also applied explainable AI techniques, such as SHAP analysis, to understand which nanoparticle features most strongly influence toxicity. This helps move beyond “black-box” predictions and provides insight into why certain nanoparticles may be more harmful than others.  The final model is available as a free web tool through the Enalos Cloud platform, allowing researchers to predict nanoparticle toxicity quickly and without the need for laboratory experiments.

Featured Image

Photo by Steve Johnson on Unsplash

Photo by Steve Johnson on Unsplash

Why is it important?

Nanomaterials are increasingly used in medicine and technology, but their safety remains a key challenge. Experimental testing alone is not sufficient to keep up with the growing number of new materials, as it is time-consuming, costly, and often lacks detailed mechanistic understanding. This work is important because it introduces a reliable, explainable, and accessible method to assess nanoparticle toxicity based on their atomic structure. By using atom-level descriptors, the approach captures fundamental properties that directly influence how nanoparticles interact with biological systems. The integration of explainable AI is particularly valuable, as it allows scientists and regulators to understand the drivers of toxicity rather than relying only on predictions. This supports safer material design (“safe-by-design”), regulatory decision-making, and more efficient risk assessment. Additionally, making the model available through a free cloud platform enables broad use by the scientific community, promoting transparency, reproducibility, and faster innovation in nanotechnology safety.

Perspectives