: Per- and polyfluoroalkyl substances (PFAS) are of concern because of their potential thyroid hormone system disruption by binding to human transthyretin (hTTR). However, the amount of experimental data is scarce. In this work, new classification and regression QSARs were developed to predict the hTTR disruption based on experimental data measured for 134 PFAS. Bootstrapping, randomization procedures, and external validation were used to check for overfitting, to avoid random correlations, and to evaluate the predictivity of the QSARs, respectively. The best QSARs were characterized by good performances (e.g., training and test accuracies in classification of 0.89 and 0.85, respectively; R2, Q2loo, and Q2F3 in regression of 0.81, 0.77, and 0.82, respectively) and significantly broader domains compared to the few existing similar models. The application of QSARs application to the OECD List of PFAS allowed for the identification of structural categories of major concern, such as per- and polyfluoroalkyl ether-based, perfluoroalkyl carbonyl, and perfluoroalkane sulfonyl compounds. Forty-nine PFAS showed a stronger binding affinity to hTTR than the natural ligand T4. Uncertainty quantification for each model and prediction further enhanced the reliability assessment of predictions. The implementation of the new QSARs in non-commercial software facilitates their application to support future research efforts and regulatory actions.
New QSAR Models to Predict Human Transthyretin Disruption by Per- and Polyfluoroalkyl Substances (PFAS): Development and Application
Evangelista M.;Chirico N.;Papa E.
2025-01-01
Abstract
: Per- and polyfluoroalkyl substances (PFAS) are of concern because of their potential thyroid hormone system disruption by binding to human transthyretin (hTTR). However, the amount of experimental data is scarce. In this work, new classification and regression QSARs were developed to predict the hTTR disruption based on experimental data measured for 134 PFAS. Bootstrapping, randomization procedures, and external validation were used to check for overfitting, to avoid random correlations, and to evaluate the predictivity of the QSARs, respectively. The best QSARs were characterized by good performances (e.g., training and test accuracies in classification of 0.89 and 0.85, respectively; R2, Q2loo, and Q2F3 in regression of 0.81, 0.77, and 0.82, respectively) and significantly broader domains compared to the few existing similar models. The application of QSARs application to the OECD List of PFAS allowed for the identification of structural categories of major concern, such as per- and polyfluoroalkyl ether-based, perfluoroalkyl carbonyl, and perfluoroalkane sulfonyl compounds. Forty-nine PFAS showed a stronger binding affinity to hTTR than the natural ligand T4. Uncertainty quantification for each model and prediction further enhanced the reliability assessment of predictions. The implementation of the new QSARs in non-commercial software facilitates their application to support future research efforts and regulatory actions.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
