Transcriptome-driven inference of adverse drug interactions

摘要 耳毒性是超过150种药物的一种使人衰弱的副作用，其中许多药物是作为多种药物的一部分处方的。 治疗包括癌症和寄生虫感染在内的广泛疾病的药物方案。不良药物- 增强耳毒性的药物相互作用（DDI）使多药物方案的实施复杂化， 特别是治疗多种并发病症。在大多数情况下，DDI目前仅在 药物已经上市，因此，有效的临床前方法来识别潜在的不良相互作用， 促进更安全的联合处方。天文数字的组合使得测量所有可能的药物 相互作用是不可行的，因此从单个化合物的数据预测耳毒性药物如何相互作用是不可行的。 必要虽然目前对特定药物类别耳毒性机制的理解， 有助于解释特定不良耳毒性DDI的临床观察结果，并有助于合理设计 候选耳保护佐剂，该策略不能预期不良耳毒性DDI或开发 用于其他较少研究的药物类别和临床开发中的一流药物的耳保护剂。到 更广泛地调查潜在的耳毒性DDI，我们将采用靛蓝（使用 chemo-Genomics and Orthology），这是一种机器学习工具，目前可以预测 在不需要特定的药物靶标信息的情况下，在多种细菌物种中的抗微生物药物活性。我们 假设我们可以利用潜在的方法来预测潜在的不利（协同）或 通过基于数据构建“INDIGO-Tox”模型，在人体中保护（拮抗）耳毒性DDI 从适当的动物系统中产生。我们将测量15种药物引起的转录谱 已知传递耳毒性或耳保护，以及相应的成对耳毒性DDI表型， 斑马鱼，一种用于研究耳毒性的成熟体内模型系统。我们将用这些数据来训练 INDIGO-Tox模型。然后，我们将使用INDIGO-Tox来预测10种其他药物之间的DDI， 斑马鱼转录组响应谱作为输入数据。我们将在斑马鱼中验证预测， 在成熟的小鼠耳毒性离体模型中转化最经验证的预测。我们还将 使用该模型来预测影响耳毒性的新基因，然后我们将在 斑马鱼成功完成将产生翻译成人类的假设，促进模型 扩展到评估可能的耳毒性相互作用，以获得更广泛的药物库，并将建立一种途径， 预测耳毒性和其他器官毒性之间的相互作用。

ABSTRACT Ototoxicity is a debilitating side effect of over 150 medications, many of which are prescribed as part of multi- drug regimens to treat a broad range of conditions including cancer and recalcitrant infections. Adverse drug- drug interactions (DDIs) that potentiate ototoxicity complicate the implementation of multi-drug regimens, particularly to treat multiple concurrent conditions. In most cases, DDIs are currently detected only after the drugs are on the market, so effective preclinical methods to identify potential adverse interactions would facilitate safer co-prescriptions. The astronomical number of combinations renders measuring all possible drug interactions infeasible, so predicting how ototoxic drugs interact from data of individual compounds is necessary. While the current understanding of mechanisms underlying ototoxicity of specific drug classes has helped to explain clinical observations of specific adverse ototoxicity DDIs, and aided rational design of candidate otoprotective adjuvants, this strategy cannot anticipate adverse ototoxicity DDIs or develop otoprotectants for other lesser studied drug classes and first-in-class drugs under clinical development. To survey more broadly for potential ototoxicity DDIs, we will adapt INDIGO (Inferring Drug Interactions using chemo-Genomics and Orthology), a machine learning tool that currently can predict synergy/antagonism of antimicrobial drug activity in multiple bacterial species without requiring specific drug target information. We hypothesize that we can harness the underlying approach to predict potentially adverse (synergistic) or protective (antagonistic) ototoxic DDIs in humans, by building an “INDIGO-Tox” model based on data generated from an appropriate animal system. We will measure transcriptional profiles elicited by 15 drugs known to convey ototoxicity or otoprotection, as well as corresponding pairwise ototoxicity DDI phenotypes in zebrafish, a well-established in vivo model system for studying ototoxicity. We will use these data to train INDIGO-Tox model. We will then use INDIGO-Tox to predict DDIs between 10 additional drugs, using their zebrafish transcriptome response profiles as input data. We will validate predictions in zebrafish, and will test translation of top validated predictions in a well-established mouse ex vivo model of ototoxicity. We will also use the model to generate predictions for novel genes that influence ototoxicity, which we will then test in zebrafish. Successful completion will generate hypotheses for translation into humans, facilitate model expansion to assessing possible ototoxic interactions for a broader library of drugs, and will establish a path to predict interactions between ototoxicity and other organ toxicities.