Gambacorta Nicola, Mastrolorito Fabrizio, Togo Maria Vittoria, Amenduni Vincenzo, Mele Marco, Liantonio Antonella, Mele Antonietta, De Luca Annamaria, Altomare Cosimo Damiano, Belgiovine Valentina, Tondo Anna Rita, Cutropia Francesca, Siragusa Lydia, Amoroso Nicola, Ciriaco Fulvio, Imbrici Paola, Trisciuzzi Daniela, Nicolotti Orazio
Division of Medical Genetics, IRCCS Foundation-Casa Sollievo della Sofferenza, San Giovanni Rotondo, Foggia, Italy; Department of Pharmacy-Pharmaceutical Sciences, University of Bari "Aldo Moro", Bari, Italy.
Department of Pharmacy-Pharmaceutical Sciences, University of Bari "Aldo Moro", Bari, Italy.
Eur J Med Chem. 2025 Jun 5;290:117575. doi: 10.1016/j.ejmech.2025.117575. Epub 2025 Mar 27.
The withdrawal of numerous approved drugs in late development stages, or even from the market, due to safety concerns remains a major challenge, contributing to the high attrition rate in drug discovery and development. Among these concerns, cardiotoxicity is a critical toxicological issue, particularly in oncology, as drugs can induce heart damage by triggering pathological conditions such as arrhythmia, myocardial infarction, and myocardial hypertrophy. Here, we introduce CUPID (Cardiotox Understanding Platform for Intelligent Drug Discovery), an explainable artificial intelligence (XAI) framework designed to predict cardiotoxicity associated with ERG (ether-à-go-go-related gene) potassium, Na1.5 sodium, and Ca1.2 calcium ion channels. The framework was trained using three carefully curated interspecies experimental datasets from the latest ChEMBL database (release 34) and the CSFP (Core-Substituent Fingerprint), which encodes molecular fragments derived from the decomposition of drug-like small molecules. By leveraging these experimental datasets, highly accurate explainable machine learning models were developed, achieving approximately 80 % accuracy in 5-fold stratified cross-validation analyses. CUPID provides a comprehensive risk assessment of early cardiotoxicity and a key feature is its interpretability: predictions are annotated with clear applicability domain information, while chemical substructures linked to cardiotoxicity risks are highlighted using SHAP (SHapley Additive exPlanations) values. This enhances molecular understanding and facilitates the rational design of safer bioactive compounds. Last but not least, CUPID is freely accessible at https://prometheus.farmacia.uniba.it/cupid.
由于安全问题，众多已获批药物在研发后期甚至从市场上撤回，这仍然是一个重大挑战，导致药物研发的高淘汰率。在这些问题中，心脏毒性是一个关键的毒理学问题，尤其是在肿瘤学领域，因为药物可通过引发心律失常、心肌梗死和心肌肥大等病理状况导致心脏损伤。在此，我们介绍CUPID（智能药物发现心脏毒性理解平台），这是一个可解释人工智能（XAI）框架，旨在预测与ERG（去极化相关基因）钾离子通道、Na1.5钠离子通道和Ca1.2钙离子通道相关的心脏毒性。该框架使用来自最新ChEMBL数据库（版本34）和CSFP（核心取代基指纹）精心策划的三个跨物种实验数据集进行训练，CSFP对源自类药物小分子分解的分子片段进行编码。通过利用这些实验数据集，开发了高度准确的可解释机器学习模型，在5折分层交叉验证分析中实现了约80%的准确率。CUPID提供早期心脏毒性的全面风险评估，其一个关键特性是可解释性：预测标注有明确的适用域信息，同时使用SHAP（Shapley值加法解释）值突出显示与心脏毒性风险相关的化学亚结构。这增强了对分子的理解，并有助于更合理地设计更安全的生物活性化合物。最后但同样重要的是，可通过https://prometheus.farmacia.uniba.it/cupid免费访问CUPID。
