“intelligent Read Across (iRA)”- A tool for read-across-based toxicity prediction of nanoparticles

The rapid advancement of nanotechnology has enabled the use of nanoparticles (NPs) in various applications, such as medicine, electrochemical sensors, and cosmetics, due to their unique physical and chemical properties. Their small size allows these particles to penetrate biological systems and interact with intracellular components, which may pose significant health risks to humans and other organisms. As a result, assessing the health risks and environmental impacts of NPs has gained considerable attention. Experimental evaluation of NP toxicity is resource-intensive and raises ethical issues; therefore, various computational methods are used for toxicity assessments. In this research, we introduce a Python-based tool called "intelligent Read Across (iRA)," which makes predictions using similarity-based read-across algorithms. In addition to toxicity endpoint predictions, this tool enables pairwise similarity calculations, read-across optimization, and the identification of important features related to read-across predictions. Similarity calculations assess how close compounds are based on their molecular descriptors. The read-across optimization feature helps determine the best hyperparameter values for the similarity measures. Furthermore, feature importance analysis evaluates the relative significance of features involved in read-across prediction. This tool has been validated using three small datasets (≤ 30 samples) containing nanotoxicity data. External validation metrics show improvements over previously reported models across all datasets. These results demonstrate the effectiveness of this similarity-based read-across method. Consequently, the developed tool can be used for accurate prediction of the toxic potential and prioritization of data-poor NPs.