Machine Learning-Based QSPR Modeling of log CMC Values of Per- and Polyfluoroalkyl Substances (PFASs) and the Identification of Property Cliffs

Per- and polyfluoroalkyl substances (PFASs) possess sufficient
water solubility and chemical durability, enabling them to affect
both terrestrial and aquatic environments. Their wide chemical diversity
makes it difficult to assess property end points experimentally. In
this study, experimental critical micelle concentration (log CMC)
data for representative PFASs and non-PFASs were used to develop computational
predictive models. We developed a quantitative structure-property
relationship (QSPR) model and a quantitative read-across structure-property
relationship (q-RASPR) model for PFASs and defluorinated/non-PFASs
separately and explored the importance of similarity and error-based
features along with molecular descriptors. Subsequently, a combined
data set of PFASs and non-PFASs was used to develop a read-across
structure property relationship coupled with the multiclass arithmetic
residuals in the K-groups analysis framework (ARKA-RASPR) for capturing
response range-specific contribution. We explored the physicochemical
insights of our developed models for PFASs and non-PFASs separately
to understand the underlying reasons for the positive or negative
correlations between the modeled descriptors and response variables.
Furthermore, we developed a Python-based expert system, PFAS_(CMC)_Predictor
v1.0 (available from https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home/pfas-cmc-predictor), to predict the CMC value of a true external set comprising PFAS
surfactants exclusively and any organic surfactants separately and
to provide applicability domain status with graphical representation,
thus emphasizing overall practicality and accessibility of this study.
Additionally, we identified the property cliffs and investigated the
reason for their cliff behavior. The developed models will further
help in mapping sustainable mitigation strategies and control measures
to reduce PFAS persistence.