Combining Systems Pharmacology Modeling With Machine Learning To Identify Sub-Populations At Risk Of Drug-Induced Torsades de Pointes

项目摘要 尖端扭转型室性心动过速是一种致命的室性心律失常，是几种常用药物的副作用。 抗心律失常药、抗生素、抗精神病药、抗组胺药和其他“非心血管”疗法。虽然 这种不良事件是罕见的，它可以导致心室颤动和心源性猝死。关于 形成这种药物诱发心律失常的高风险与低风险之间的潜在差异 阻止了在预防它方面取得的任何重大进展。与其简单地将这些药物从市场上移除， 对从治疗中受益的患者的生理和临床特征进行更仔细的检查， 形成心律不齐的患者需要进行手术。这突出了精准医疗的理念， 确定可能从治疗中获益的患者相关亚组与 对药物引起的不良事件高度敏感。目前预测风险的标准， 已经证明，细胞的动作电位（AP）持续时间和超声心动图（ECG）上的QT间期延长 无效。因此，需要从细胞和组织水平提取相关信息， 施用治疗剂以检测仅在高风险人群中明显的模式。分析这个 概念，我计划（1）在机械水平上解释健康和高危患者之间的差异，（2） 识别可以早期预测风险的重要AP和ECG特征，以及（3）将生理和 临床发现，以改善高风险人群的概况和描述。我将联合收割机 补充计算技术：（1）机械定量系统药理学模拟 心脏细胞和组织的模型;（2）先进的机器学习方法，可以识别隐藏的 模式.因此，本项目旨在开发一种算法，这将提高风险预测和升级 目前不完善和不可靠的预防性治疗处方标准。

Project Summary Torsades de Pointes, a lethal ventricular arrhythmia, is a side effect of several commonly used antiarrhythmics, antibiotics, antipsychotics, antihistamines and other ‘non-cardiovascular’ therapies. Though this adverse event is rare, it can lead to ventricular fibrillation and sudden cardiac death. The ignorance about the underlying differences between those at high risk versus low risk of forming this drug-induced arrhythmia halts any considerable progress in preventing it. Rather than simply removing these drugs from the market, a closer examination of the physiological and clinical traits of patients who benefited from the treatment and those who formed the arrhythmia needs to be performed. This highlights the idea of precision medicine and the importance of identifying relevant sub-groups of patients likely to benefit from a treatment versus those who are highly susceptible to a drug-induced adverse event. The current standards for predicting risk, a lengthened action potential (AP) duration of cells and a prolonged QT interval on an echocardiogram (ECG) have proven ineffective. Thus, there is a need to extract pertinent information from the cellular and tissue levels before administration of the therapeutic to detect patterns only apparent in the high-risk population. To analyze this concept, I plan to (1) explain at a mechanistic level the differences between the healthy and at-risk patients, (2) identify important AP and ECG signatures that can predict risk early on, and (3) connect the physiological and clinical findings to improve the profile and description of the high-risk population. I will combine two complementary computational techniques: (1) simulations with mechanistic quantitative systems pharmacology models of heart cells and tissues; and (2) advanced machine learning approaches that can identify hidden patterns. Thus, this project aims to develop an algorithm which will improve risk prediction and upgrade the current imperfect and unreliable standards for prescribing proarrhythmic therapies.