ChemNTP:\nAdvanced Prediction of Neurotoxicity Targets\nfor Environmental Chemicals Using a Siamese Neural Network

Environmental\nchemicals can enter the human body through various\nexposure pathways, potentially leading to neurotoxic effects that\npose significant health risks. Many such chemicals have been identified\nas neurotoxic, but the molecular mechanisms underlying their toxicity,\nincluding specific binding targets, remain unclear. To address this,\nwe developed ChemNTP, a predictive model for identifying neurotoxicity\ntargets of environmental chemicals. ChemNTP integrates a comprehensive\nrepresentation of chemical structures and biological targets, improving\nupon traditional methods that are limited to single targets and mechanisms.\nBy leveraging these structural representations, ChemNTP enables rapid\nscreening across 199 potential neurotoxic targets or key molecular\ninitiating events (MIEs). The model demonstrates robust predictive\nperformance, achieving an area under the receiver operating characteristic\ncurve (<i>AUC</i><sub>ROC</sub>) of 0.923 on the test set.\nAdditionally, ChemNTP’s attention mechanism highlights critical\nresidues in binding targets and key functional groups or atoms in\nmolecules, offering insights into the structural basis of interactions.\nExperimental validation through <i>in vitro</i> enzyme activity\nassays and molecular docking confirmed the binding of eight polybrominated\ndiphenyl ethers (PBDEs) to acetylcholinesterase (AChE). We also provide\na user-friendly software interface to facilitate the rapid identification\nof neurotoxicity targets for emerging environmental pollutants, with\npotential applications in studying MIEs for more types of toxicity.