Towards higher scientific validity and regulatory acceptance of predictive models for PFAS

Abstract

Per- and polyfluoroalkyl substances (PFAS) are widespread in the environment. Properly developed QSAR/QSPR models can be used to assess the impact of these chemicals on humans and the environment. This work assesses 38 in silico models developed for this group of compounds, which mainly show physicochemical (22), and also toxic (8) and ecotoxic (8) properties. The evaluation of the models was carried out based on the (Q)SAR Model Reporting Format (QMRF), which was found in the QSAR Database (5) or was prepared manually, according to the information contained in scientific publications based on the QMRFEditor-v3.0.0 format (33). We based our evaluations on an individual assessment of each of the OECD principles described in the document and then summing up everything together. During the analysis, we identified 22 models as scientifically valid and could be used in the prediction of new compounds. Twelve of them contained all the information necessary to reproduce the model, and another 10, despite the lack of some information, are still reproducible. The other 16 models do not contain enough information to reproduce them and therefore they are scientifically invalid. The present work allows identifying the remaining gaps, needs, and recommendations that should be considered in further development of predictive models in the PFAS area.