Abstract :Inorganic nanomaterials have been increasing attraction of science and technology research since their discovery. Due to their unique physical and chemical properties and ease of their synthesis, they have great potential on many application areas including energy, electronics, photonics, biomedical devices, sensing and biosensing [1-3]. The use of the inorganic nanoparticles (NPs) in biomedical applications has tremendous importance for last two decades especially in non-invasive and long-term imaging of detection and treatment of many diseases such as cancer, infectious diseases etc.[4]. However, controlling the cytotoxicity effect of the inorganic NPs is an unresolved issue in non-invasive bioimaging studies. Currently, there are still many obstacles to solve this issue. Furthermore, the common experiments reported in literature are mostly based on cultured cells, yet these processes are time consuming and costly. In order to reduce these costs, the cytotoxicity prediction of the inorganic NPs using machine-learning (ML) method can provide an alternative approach. ML that is a subset of an artificial intelligence [5-6] has used to reduce the human work. ML provides algorithms that are more efficient when subject to relevant data than giving instructions [7]. Decision trees (DT) is one of the most popular method for both classification and regression in ML. In this method, any path starts from the root and continues until it reaches a Boolean outcome at the leaf node [8-11]. In this study, the cytotoxicity prediction of the inorganic NPs is studied based on DT algorithms including Fine trees, Medium trees, Coarse trees, Boosted and Bagged trees methods. The results related to the dataset in [12] are given in Table 1. Table 1: Regression results according to the DT algorithms Method RMSE R-Squared MSE Fine Tree 0.23305 0.56 0.054313 Medium Tree 0.24643 0.51 0.060727 Coarse Tree 0.2415 0.53 0.058325 Boosted Trees 0.1932 0.70 0.037328 Bagged Trees 0.19032 0.71 0.036221 As seen in Table 1, Bagged trees method is the best method in terms of mean absolute error (RMSE= 0.19032), mean square error (MSE=0.036221), and R-squared measure (R-squared=0.71) among all the configurations. Keywords : Nanomaterials, toxicity, OCHEM, machine-learning
