Predicting adverse drug reactions via networks of drug binding pocket similarity

项目总结 疾病预防控制中心估计，药物不良反应(ADR)每年导致130万急诊科就诊 在美国，有数十万这样的患者需要住院治疗。不良反应通常由以下原因引起 药物结合体内的蛋白质，而不是预期的目标。预测这种脱离目标的绑定是困难的。 有一些方法可以使用3D分子结构来预测小分子是否可以结合给定的蛋白质，但 人类蛋白质组的大部分没有实验解决的结构。最近的突破 在蛋白质结构预测中实现了对几乎任何蛋白质结构的高置信度预测 仅序列，意味着我们可以利用整个人类蛋白质组的结构信息在某种程度上 这在两年前是不可能的。此外，结构信息学算法的最新进展 提高了我们识别具有高结合倾向的蛋白质表面位置的能力；尽管如此， 目前的ADR预测算法不能同时利用功能上的结合信息 并做出可解释的预测，以指导药物设计。我提议创造一个 方法在蛋白质组水平上以可解释的方式预测药物结合袋和不良反应。这就做 通过1)建立已知和预测的药物袋子对的图形表示；2)使用 用于评估与口袋和药物相关的ADR的图表；以及3)扩展口袋和ADR预测 方法预测和解释蛋白质组非靶点结合引起的不良反应。建议的适用范围 对整个人类蛋白质组的方法将允许预测药物的潜在ADR，然后将其用于 人类，改善药物开发，减少不良反应的发生。 我将在斯坦福大学的Russ Altman博士的实验室进行这个项目，在那里我正在研究我的 长期的职业目标是成为一名开发计算方法的独立研究员 加速药物开发，帮助在分子水平上了解药物反应。我的训练 环境为我实现这个目标做了很好的准备，因为奥特曼博士在指导方面有很好的记录 研究生和斯坦福大学提供了过多的教育资源和高度 协作研究环境。Altman团队已经开发出表征蛋白质的算法 微环境，在计算结构生物学和药物反应研究方面都有历史， 为我提供了与我提议的工作高度相关的领域的专家。超越了 建议的研究，我的培训计划包括参加研讨会和会议，与其他 研究小组，采取额外的课程作业，教学，以及口头和书面交流我的工作。

PROJECT SUMMARY The CDC estimates that adverse drug reactions (ADRs) cause 1.3 million emergency department visits annually in the U.S., and that hundreds of thousands of these patients require hospitalization. ADRs are often caused by drugs binding proteins in the body that were not intended targets. Predicting this off-target binding is difficult. There are methods that use 3D molecular structure to predict if a small molecule can bind a given protein, but the majority of the human proteome does not have an experimentally-solved structure. Recent breakthroughs in protein structure prediction have enabled high confidence prediction of nearly any protein's structure from sequence alone, meaning that we can leverage structure information for the entire human proteome in a way that was impossible two years ago. Additionally, recent advances in structural informatics algorithms have improved our ability to identify locations on a protein surface with high binding propensity; despite this, current ADR prediction algorithms are unable to both leverage binding information of functionally uncharacterized proteins and make interpretable predictions that can guide drug design. I propose to create methods to predict drug binding pockets and ADRs in an interpretable manner at the proteome scale. I will accomplish this by 1) building a graph representation of known and predicted drug-pocket pairs; 2) using this graph to estimate ADRs associated with pockets and drugs; and 3) extending the pocket and ADR prediction methods to predict and explain ADRs caused by proteome-wide off-target binding. Application of the proposed method to the entire human proteome will allow the prediction of a drug's potential ADRs before it is used in humans, improving drug development and reducing the number of ADRs experienced. I will conduct this project in the lab of Dr. Russ Altman at Stanford University, where I am working toward my long-term career goal of becoming an independent researcher developing computational methods that accelerate drug development and aid understanding of drug response at the molecular level. My training environment sets me up well to achieve this goal as Dr. Altman has an excellent track record of mentoring graduate students and Stanford University provides a plethora of educational resources and a highly collaborative research environment. The Altman group has developed algorithms for characterizing protein microenvironments and has a history in both computational structural biology and drug response research, providing me with easy access to experts in domains highly relevant to my proposed work. Beyond the proposed research, my training plan includes attending seminars and conferences, collaborating with other research groups, taking additional coursework, teaching, and oral and written communication of my work.