Development of a Novel In Silico Classification Model to Assess Reactive Metabolite Formation in the Cysteine Trapping Assay and Investigation of Important Substructures

• Published in: Biomolecules (Basel, Switzerland) Vol. 14; no. 5; p. 535
• Main Authors: Umemori, Yuki, Handa, Koichi, Yoshimura, Saki, Kageyama, Michiharu, Iijima, Takeshi
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 30-04-2024
• Subjects: Algorithms; Assaying; Chemical and Drug Induced Liver Injury - diagnosis; Chemical and Drug Induced Liver Injury - metabolism; Classification; Clinical trials; Computer Simulation; Cysteine; Cysteine - metabolism; cysteine trapping assay; Drug dosages; drug-induced liver injury (DILI); Evaluation; Gallbladder diseases; hepatotoxicity; Humans; idiosyncratic DILI; Machine Learning; message passing neural network; Metabolites; Methods; Microsomes; Microsomes, Liver - metabolism; Neural networks; Neural Networks, Computer; Pharmaceutical industry; Physiological aspects; Proteins; random forest; Sensors; Toxicity; Trapping; Japan; United States > US; Basel Switzerland; Switzerland; toxicity; substructures; message passing neural network; QSAR; hepatotoxicity; idiosyncratic DILI; drug-induced liver injury (DILI); random forest; reactive metabolite; cysteine trapping assay
• ISSN: 2218-273X; 2218-273X
• DOI: 10.3390/biom14050535

Predicting whether a compound can cause drug-induced liver injury (DILI) is difficult due to the complexity of drug mechanism. The cysteine trapping assay is a method for detecting reactive metabolites that bind to microsomes covalently. However, it is cumbersome to use 35S isotope-labeled cysteine for this assay. Therefore, we constructed an in silico classification model for predicting a positive/negative outcome in the cysteine trapping assay. We collected 475 compounds (436 in-house compounds and 39 publicly available drugs) based on experimental data performed in this study, and the composition of the results showed 248 positives and 227 negatives. Using a Message Passing Neural Network (MPNN) and Random Forest (RF) with extended connectivity fingerprint (ECFP) 4, we built machine learning models to predict the covalent binding risk of compounds. In the time-split dataset, AUC-ROC of MPNN and RF were 0.625 and 0.559 in the hold-out test, restrictively. This result suggests that the MPNN model has a higher predictivity than RF in the time-split dataset. Hence, we conclude that the in silico MPNN classification model for the cysteine trapping assay has a better predictive power. Furthermore, most of the substructures that contributed positively to the cysteine trapping assay were consistent with previous results.