Perspectives from the NanoSafety Modelling Cluster on the validation criteria for (Q)SAR models used in nanotechnology

Abstract

Nanotechnology and the production of nanomaterials have been expanding rapidly in recent years. Since many types of engineered nanoparticles are suspected to be toxic to living organisms and to have a negative impact on the environment, the process of designing new nanoparticles and their applications must be accompanied by a thorough risk analysis. (Quantitative) Structure-Activity Relationship ([Q]SAR) modelling creates promising options among the available methods for the risk assessment. These in silico models can be used to predict a variety of properties, including the toxicity of newly designed nanoparticles. However, (Q)SAR models must be appropriately validated to ensure the clarity, consistency and reliability of predictions. This paper is a joint initiative from recently completed European research projects focused on developing (Q)SAR methodology for nanomaterials. The aim was to interpret and expand the guidance for the well-known "OECD Principles for the Validation, for Regulatory Purposes, of (Q)SAR Models", with reference to nano-(Q)SAR, and present our opinions on the criteria to be fulfilled for models developed for nanoparticles.

Practices and Trends of Machine Learning Application in Nanotoxicology

“…In addition to descriptors' reduction, a feature selection process is followed in order to optimize the performance of the model. Feature selection may be appropriate for two key reasons including to avoid overfitting training data and, second, to enable expert assessment of the mechanistic basis for the model [4,24]. Almost half of the identified studies applied some form of a feature selection process to their initial dataset.…”

“…The methodology used by Puzyn et al [35] added to the validation set some cases, which do not fall in the range of the training set (validation and reliability testing at the same time). The complete dataset should be provided to potential dataset users, including nanomaterial, endpoints, and descriptor information, together with the clearly defined training and test sets [4].…”

“…GA showed great performance among the different methods for feature selection, while ERM was superior in some cases as a total search algorithm and, thus, less reliant on the initial set of descriptors. Such cases make the selection of statistical features a dynamic research area [4]. We recommend either a combination of different feature selection techniques to evaluate possible differences in the results, or, more efficiently, an integration of techniques proven to outperform individual methods and mitigate any method bias [143].…”

“…The latter can be covered by using the PCA score plot or diversity analysis [46,55] and investigation of multiple splitting can be performed following the methodology from Puzyn et al [35], which ensures that validation data are evenly distributed within the range of toxicity of the training dataset. The complete dataset of substances, endpoints, and descriptor values should be annexed in each analysis, along with clearly defined learning and test sets [4].…”

Validation of Computational Methods Applied in Molecular Modeling of Caffeine With Epithelial Anticancer Activity: Theoretical Study of Geometric, Thermochemical and Spectrometric Data

“…3 Validation of methods and models has been used in several areas of the knowledge, in special for in silico determinations, and the strategy for such use depends on what the scientist aims to predict, such as example the assessment of the toxicity of nanomaterials using Quantitative Structure-Activity Relationships (QSAR) models, then validated in agreement with the principles of the OECD (Organization for Economic Co-operation and Development) to ensure clarity, consistency and reliability of model forecasts for regulatory purposes. 4 Paramagnetic NMR techniques and computational modeling approaches for analyses of molecular structures and receptor-ligand complexes have been used as alternatives to reduce the difficulty of obtaining structural data from biological systems because they are complex, dynamic and with high degree of disorder. However, they point out that 3D modeling, validated from paramagnetic NMR experimental data, is an important tool useful for understanding mechanism, recognition and biomolecular function of complex biological systems.…”