P07-20

Constructing a machine learning model for discriminating Urotensin-II receptor inhibitors and its application

Kentaro KAWAI *¹, Momoko KYUTA², Runa MINATO², Shoki HOSHIKAWA¹, Reiko KONISHI², Kazuyuki SATO¹, Kohji KOMORI², Ko KAWADA², Akira MUKAI²

¹Laboratory for Medicinal Chemistry, Faculty of Pharmaceutical Sciences, Setsunan University
²Laboratory for Clinical Pharmacology and Therapeutics, Faculty of Pharmaceutical Sciences, Setsunan University
( * E-mail: kentaro.kawai@pharm.setsunan.ac.jp )

Rare adverse reactions of cardiotoxicity have been reported with Lemdecivir (RDV), a drug approved for COVID-19. Although the detailed mechanism of occurrence was unknown, it was reported that the Urotensin-II receptor (UT2R) pathway is involved in RDV cardiotoxicity and UT2R inhibition reduces the occurrence of cardiotoxicity. We have therefore conducted an analysis of US Food and Drug Administration (FDA)-approved drugs using the machine-learning model for UT2R activity to derive new candidate compounds for the treatment of RDV-induced cardiovascular events.
Compounds acting on UT2R were extracted from the ChEMBL database. Compounds with a pChEMBL value of 6 or higher were labeled as active. However, as there were few compounds with a pChEMBL value of less than 6, a random sampling of data from among the ChEMBL listed compounds was used to select compounds as inactive compounds. Each compound was converted to six fingerprints, which were trained using four classification models, including SVM and random forests. The created models were then used to discriminate FDA-approved drugs, and compounds with the potential to show binding properties to UT2R were obtained.
The ROC-AUC and Cohen's Kappa of 24 models combining machine learning algorithms and fingerprints showed high values. The FDA-approved drugs were analyzed, and drugs that were suggested to bind to UT2R were found. FAERS was used to analyze cardiotoxicity-related conditions, and the relationship between the drugs and adverse events such as arrhythmia was investigated.