Application of machine learning to predict the inhibitory activity of organic chemicals on thyroid stimulating hormone receptor

CatNet: Sequence-based deep learning with cross-attention mechanism for identifying endocrine-disrupting chemicals

Endocrine-disrupting chemicals (EDCs) pose significant environmental and health risks due to their potential to interfere with nuclear receptors (NRs), key regulators of physiological processes. Despite the evident risks, the majority of existing research narrows its focus on the interaction between compounds and the individual NR target, neglecting a comprehensive assessment across the entire NR family. In response, this study assembled a comprehensive human NR dataset, capturing 49,244 interactions between 35,467 unique compounds and 42 NRs. We introduced a cross-attention network framework, "CatNet", innovatively integrating compound and protein representations through cross-attention mechanisms. The results showed that CatNet model achieved excellent performance with an area under the receiver operating characteristic curve (AUCROC) = 0.916 on the test set, and exhibited reliable generalization on unseen compound-NR pairs. A distinguishing feature of our research is its capacity to expand to novel targets. Beyond its predictive accuracy, CatNet offers a valuable mechanistic perspective on compound-NR interactions through feature visualization. Augmenting the utility of our research, we have also developed a graphical user interface, empowering researchers to predict chemical binding to diverse NRs. Our model enables the prediction of human NR-related EDCs and shows the potential to identify EDCs related to other targets.

A review on structural mechanisms of protein-persistent organic pollutant (POP) interactions

With the advent of the industrial revolution, the accumulation of persistent organic pollutants (POPs) in the environment has become ubiquitous. POPs are halogen-containing organic molecules that accumulate, and remain in the environment for a long time, thus causing toxic effects in living organisms. POPs exhibit a high affinity towards biological macromolecules such as nucleic acids, proteins and lipids, causing genotoxicity and impairment of homeostasis in living organisms. Proteins are essential members of the biological assembly, as they stipulate all necessary processes for the survival of an organism. Owing to their stereochemical features, POPs and their metabolites form energetically favourable complexes with proteins, as supported by biological and dose-dependent toxicological studies. Although individual studies have reported the biological aspects of protein-POP interactions, no comprehensive study summarizing the structural mechanisms, thermodynamics and kinetics of protein-POP complexes is available. The current review identifies and classifies protein-POP interaction according to the structural and functional basis of proteins into five major protein targets, including digestive and other enzymes, serum proteins, transcription factors, transporters, and G-protein coupled receptors. Further, analysis detailing the molecular interactions and structural mechanism evidenced that H-bonds, van der Waals, and hydrophobic interactions essentially mediate the formation of protein-POP complexes. Moreover, interaction of POPs alters the protein conformation through kinetic and thermodynamic processes like competitive inhibition and allostery to modulate the cellular signalling processes, resulting in various pathological conditions such as cancers and inflammations. In summary, the review provides a comprehensive insight into the critical structural/molecular aspects of protein-POP interactions.

Identification of active or inactive agonists of tumor suppressor protein based on Tox21 library

Exposure of cells to xenobiotic human-made products can lead to genotoxicity and cause DNA damage. It is an urgent need to quickly identify the chemicals that cause DNA damage, and their toxicity should be predicted. In this study, recursive partitioning (RP), binary logistic regression, and one machine learning approach, namely, random forest (RF) classifier, were used to predict the active and inactive compounds of a total 5036 data based on the assay conducted by a β-lactamase reporter gene under control of the p53 response element (p53RE) from Tox21 library. Results show that the binary logistic regression model with a threshold of 0.5 has a high accuracy rate (83%) to distinguish active and inactive compounds. The RF classifier method has satisfactory results, with an accuracy rate (84.38%) approximately higher than that of binary logistic regression. The models established can identify compounds that induce DNA damage and activate p53, and provide a scientific basis for the risk assessment of organic chemicals in the environment.