Development and Experimental Validation of Regularized Machine Learning Models Detecting New, Structurally Distinct Activators of PXR

Abstract

The pregnane X receptor (PXR) regulates the metabolism of many xenobiotic and endobiotic substances. In consequence, PXR decreases the efficacy of many small-molecule drugs and induces drug-drug interactions. The prediction of PXR activators with theoretical approaches such as machine learning (ML) proves challenging due to the ligand promiscuity of PXR, which is related to its large and flexible binding pocket. In this work we demonstrate, by the example of random forest models and support vector machines, that classifiers generated following classical training procedures often fail to predict PXR activity for compounds that are dissimilar from those in the training set. We present a novel regularization technique that penalizes the gap between a model’s training and validation performance. On a challenging test set, this technique led to improvements in Matthew correlation coefficients (MCCs) by up to 0.21. Using these regularized ML models, we selected 31 compounds that are structurally distinct from known PXR ligands for experimental validation. Twelve of them were confirmed as active in the cellular PXR ligand-binding domain assembly assay and more hits were identified during follow-up studies. Comprehensive analysis of key features of PXR biology conducted for three representative hits confirmed their ability to activate the PXR.

Integrated Approach for Testing and Assessment for Developmental Neurotoxicity (DNT) to Prioritize Aromatic Organophosphorus Flame Retardants

“…TDCIPP was active in several of the assays in ICE with assays for gene expression regulation of the Pregnane-X receptor (PXR) being most sensitive at 10-20-fold lower concentrations than the DNT battery. PXR is a key regulator in the xenobiotic response pathway as it regulates the expression of various drug metabolizing enzymes, such as P450s and glutathione S-transferases [71]. However, activation of xenobiotic nuclear receptors, including PXR, has also been identified as an MIE in AOP8 (https://aopwiki.org/aops/8) (access date 15 April 2024) leading to thyroid hormone disruption that consequently can result in altered neurodevelopment.…”

Allosteric Antagonism of the Pregnane X Receptor (PXR): Current-State-of-the-Art and Prediction of Novel Allosteric Sites

“…Artificial intelligence (AI) is the fastest-growing technology in the life sciences and drug discovery. Computational docking, molecular dynamics simulations, and machine learning approaches are useful for designing and discovering new chemical entities with nuclear receptors [101,102]. Support vector machine algorithms (SVM) and pocket-based analysis have been used to predict allosteric sites in proteins.…”

Machine learning based models for high-throughput classification of human pregnane X receptor activators

“…Ligand-based computational models employing pharmacophore mapping and quantitative structure activity relationships (QSARs) have previously been developed to discriminate PXR activators from non-activators 5,[8][9][10][11][12][13][14][15] . Till now, several QSAR models have been built using machine learning methods including k-NN, naïve Bayesian, probabilistic neural networks, artificial neural networks and random forest [16][17][18][19][20][21][22] . The lack of larger PXR datasets has restricted PXR classification models applying to a large scale of compounds screening.…”

“…5,[8][9][10][11][12][13][14][15] To date, several QSAR models have been built using machine learning methods, including k-NN, Naïve Bayesian, probabilistic neural networks, articial neural networks, and random forest. [16][17][18][19][20][21][22] The lack of larger PXR datasets has restricted the application of PXR classication models to a large scale of compound screening. Therefore, a comprehensive and large dataset is required to develop machine learning driven models with a broader chemical space and higher generalization ability.…”