Machine learning-accelerated determination of immunotoxicities of liquid crystal monomers

Abstract

Liquid crystal monomers (LCMs), essential components of liquid crystal displays (LCDs), have emerged as a growing global concern because of their potential adverse effects on human health. However, experimental evidence elucidating their toxicological mechanisms remains scarce. In this study, a novel mode-of-action (MOA)-based graph neural network (GNN) framework was introduced to (1) identify the molecular initiating event (MIE) through which LCMs disrupt immune homeostasis, (2) predict their binding affinity to 11β-hydroxysteroid dehydrogenase type 1 (HSD11B1), and (3) identify toxicity-relevant substructures to guide the molecular design of safer LCMs. First, an unsupervised clustering approach was applied to categorize over 1,400 LCMs into four structural clusters. Then, two virtual screening methods (i.e., 2D/3D structural similarity analysis and a pharmacophore-based model) were applied to identify 121 possible receptors that could interact by potential insertions of LCM with a total universe of approximately 2,000 human protein receptors. Through “fingerprint”-based, t-distributed stochastic neighbor embedding (t-SNE) and molecular similarity analysis using PubChem, LCMs were found to be potential inhibitors of HSD11B1. Third, docking simulations revealed that highly fluorinated and ester-containing monomers exhibit stronger binding affinities at the HSD11B1 active site, primarily through stable hydrogen bonding. Finally, a GNN model was developed, which accurately predicted LCM-HSD11B1 binding affinities (R² = 0.90, RMSE = 2.36) and identified key structural features contributing to immunotoxicity. Furthermore, a user-friendly, web-based prediction tool was developed to facilitate broader applications. These findings reveal HSD11B1 as a novel MIE of LCM-induced immunotoxicity and provide a practical immunotoxicity prediction platform to support risk assessment and the design of environmentally friendly LCM alternatives in LCD manufacturing.