Fig. 10

Classification tree (CT) model for cytotoxic responses in the CFE assay for human lung epithelial A549 cells. The CT model performs recursive binary splitting, meaning each node divides the data into two subsets based on a single feature. At each internal node, a decision rule is applied: if the condition is met (the answer is “yes”), the data is directed to the left child node; otherwise (the answer is “no”), the data is directed to the right child node. For example, at the root node, the decision rule ‘ZP DMEM T0 < 0.39’ means that the left branch corresponds to data (i.e., NMs) that satisfies this condition, while the right branch represents data (i.e., NMs) where ‘ZP DMEM T0’ is greater than or equal to 0.39. Toxic nanomaterials are marked in red, slightly toxic in grey, and non-toxic in green. Each node displays: (1) the predicted toxicity class (toxic, slightly toxic, or non-toxic), (2) the predicted probability for each class, and (3) the percentage of observations within the node. It is important to note that the CT model utilizes auto-scaled feature values. Therefore, a ‘ZP DMEM T0’ value of 0.39 corresponds to − 8.329 (mV), a ‘HDD DMEM T0’ value of -0.18 corresponds to 247.32 (nm), and ‘Aspect ratio’ values of − 0.21 and − 0.18 correspond to 1.652 and 2.789, respectively. It is also worth noting that a single feature may appear multiple times within a given CT model