Machine Learning-Guided Engineering of Protease Modulators

项目摘要 调节蛋白水解酶的活性是治疗癌症、自身免疫和感染的中心策略。然而， 针对蛋白酶(小分子和抗体)的选择性和有效治疗药物的发现和设计在很大程度上依赖于 关于低效的、重复的过程。因此，开发一种蛋白酶药物需要几年的时间，即使到那时，大多数蛋白酶 药物是活性部位的抑制剂，在区分相关的蛋白酶时往往存在选择性低的问题。由于 蛋白水解性失调的复杂性，恢复动态平衡不仅需要选择性的抑制剂，还需要 可以重新编程蛋白酶的选择性。不幸的是，目前还没有基于系统化的、 和量化设计原则。 为了应对这些挑战，该提案寻求首次将机器学习工具结合起来，下一步- 世代DNA测序，以及基于酵母的高通量功能筛选，以加速分离和 纳米体基蛋白水解酶调节剂的设计。功能选择将执行两个任务：(I)选择纳米体 来自基于所需功能的合成文库，以及(Ii)关联配体：表位相互作用与功能结果。 这些实验将生成高质量的数据集，这些数据集将训练机器学习算法(ML)来预测 仅基于其序列特征的基于纳米体的调节剂的效力、选择性和机制。 这种机器学习辅助的策略将加速发现稀有而有效的蛋白酶调节剂和 绕过基于结构的方法的限制。此外，精选的蛋白酶调节纳米体序列的数据集 将为未来的实验和硅胶运动提供参考和设计指南。这项工作意义重大 对生物医学研究和公共卫生感兴趣，包括精选的蛋白酶，如丙型肝炎病毒蛋白酶，MMPs， 跨膜丝氨酸蛋白酶2(新冠肺炎)、β分泌酶和胰岛素降解酶。此外，拟议的研究 为回答基本的生化问题提供基础，这些问题是关于合成配体如何映射和调节 蛋白水解酶和其他蛋白质修饰酶的功能格局。

Project Summary Modulating the activity of proteases is a central strategy for treating cancer, autoimmunity, and infection. However, the discovery and design of selective and potent therapeutics targeting proteases (small-molecules and antibodies) largely rely on inefficient, iterative processes. As a result, it takes several years to develop a protease drug, and even then, most protease drugs are active site inhibitors that often suffer from low selectivity in distinguishing related proteases. Due to the complexity of proteolytic dysregulation, restoring homeostasis requires not only selective inhibitors but also ligands that can reprogram protease selectivity. Unfortunately, no platform exists to engineer protease modulators based on systematic, and quantitative design principles. To address these challenges, this proposal seeks to combine for the first time Machine Learning tools, Next- Generation DNA sequencing, and a yeast-based high-throughput functional screen to accelerate the isolation and design of nanobody-based protease modulators. The functional selection will perform two tasks: (i) select nanobodies from synthetic libraries based on a desired function and (ii) correlate ligand: epitope interactions to a functional outcome. These experiments will generate high-quality datasets that will train machine learning algorithms (ML) to predict the potency, selectivity, and mechanisms of nanobody-based modulators based on their sequence features alone. This machine learning-aided strategy will accelerate the discovery of rare and potent protease modulators and bypass the limitations of structure-based methods. Moreover, curated datasets of protease modulatory nanobody sequences will provide reference and design guidelines for future experimental and in silico campaigns. This work is of significant interest to biomedical research and public health and includes select proteases such as Hepatitis C virus protease, MMPs, transmembrane serine protease 2 (COVID-19), β-secretase, and insulin-degrading enzyme. Moreover, the proposed studies provide a foundation to answering fundamental biochemical questions on how synthetic ligands can map and modulate the functional landscape of proteases and other protein-modifying enzymes.