Decision tree models to classify nanomaterials according to the DF4nanoGrouping scheme

Abstract

To keep pace with its rapid development an efficient approach for the risk assessment of nanomaterials is needed. Grouping concepts as developed for chemicals are now being explored for its applicability to nanomaterials. One of the recently proposed grouping systems is DF4nanoGrouping scheme. In this study, we have developed three structure-activity relationship classification tree models to be used for supporting this system by identifying structural features of nanomaterials mainly responsible for the surface activity. We used data from 19 nanomaterials that were synthesized and characterized extensively in previous studies. Subsets of these materials have been used in other studies (short-term inhalation, protein carbonylation, and intrinsic oxidative potential), resulting in a unique data set for modeling. Out of a large set of 285 possible descriptors, we have demonstrated that only three descriptors (size, specific surface area, and the quantum-mechanical calculated property 'lowest unoccupied molecular orbital') need to be used to predict the endpoints investigated. The maximum number of descriptors that were finally selected by the classification trees (CT) was very low-one for intrinsic oxidative potential, two for protein carbonylation, and three for NOAEC. This suggests that the models were wellconstructed and not over-fitted. The outcome of various statistical measures and the applicability domains of our models further indicate their robustness. Therefore, we conclude that CT can be a useful tool within the DF4nanoGrouping scheme that has been proposed before. ARTICLE HISTORY

Towards an alternative to nano-QSAR for nanoparticle toxicity ranking in case of small datasets

“…And as a matter of fact in silico models are the subject of intensive research and different computational models for nanomaterials have been reported (Lamon et al 2018). Among them, QSAR models (Fourches et al 2010;Puzyn et al 2011;Winkler et al 2013;Gajewicz et al 2015b;Pan et al 2016), read-across (Gajewicz et al 2015a(Gajewicz et al , 2017Gajewicz 2017a, b), neural network (Fjodorova et al 2017) or decision tree (Gajewicz et al 2018) classifications. In the present study, we could not reasonably build a QSAR model because our dataset was too small.…”

“…For further experiment, results provided using decisions trees can be used as a decision aid tool to decide on the most important features to measure experimentally. Also, the leave-one-out approach prevents the bias related to isolated training and test experiments and allows an objective validation of the provided decision aid tool as presented by Gajewicz et al (2018). Decision tree classification appears as complementary to partial least square regression analysis as this latter allows a very binary classification of the nanoparticles (i.e.…”

“…We may therefore wonder if alternative mathematical models, less complex and less sophisticated cannot be sufficient to rank, although more modestly but still efficiently, nanomaterials with respect to their toxicity in case of small datasets. And indeed, recently, other approaches have gained interest such as classification trees or read-across basedapproaches (Gajewicz 2017a, b;Gajewicz et al 2018). In this context, the aim of the present study was to determine if tools as simple as a simple partial least square regression or decision trees could accurately rank nanoparticles (i.e.…”

Practices and Trends of Machine Learning Application in Nanotoxicology

“…CT starts with a 'root node' that contains all objects (i.e., NMs), and then divides by recursive binary splitting into child nodes. Each split is defined by a threshold that takes into account the selected descriptor values at a given stage [105]. The GPTree uses a simplified fitness function from a random population of solutions with repeated attempts to find better solutions through the application of genetic operators.…”

“…Another method is the multiple threshold method used by Chau and Yap [121], which is a method originally proposed by G. Fumera [139]. The AD can also be calculated by the standardization approach, which is a straightforward method proposed by Roy et al [140] for terming the outliers and for identifying compounds outside the domain (validation and prediction set) [105]. Compared with the leverage strategy, the proposed method works well.…”