Informatics and Machine Learning Modules for Research Planning, Scheduling, Simulation, and Optimization in the ASPIRE Autonomous Laboratory

项目总结 获得复杂的化学物质(例如，小分子候选药物)是一项核心要求 用于检验生物学假说和探测人类健康。当前的化学方法 综合依赖于耗时的规划和劳动密集型的人工综合，这是一种 发现新的功能分子中的限速步骤。此协作项目 包括多个虚拟模块的开发，以支持多步骤化工 在自主实验室中合成新分子。这些模块旨在 使传统的合成化学家除了使用集成的 NCATS的ASPIRE团队正在开发硬件平台。计算机辅助综合 规划可以被视为一个分层的精化过程，从以下列表开始 感兴趣的分子：(1)逆合成计划，以确定合适的起始材料和 中间体，(2)反应条件推荐，以确定每个中间体 应运行反应步骤，(3)将假设的反应步骤转换为动作序列 可在自动化硬件上执行。可选但有价值的组件包括(4)录制 通过程序实验规划模块，(5)优化时序 最有效地合成多个合成目标的动作序列 (6)基于实验的工艺参数迭代优化 反馈循环中的响应。该计划将通过以下方式满足这些需求 在图网络中使用最先进的算法开发新的软件解决方案 理论、化学信息学、化学深度学习和最优化。软件模块将 使用已建立的软件开发最佳实践编写，以便于跨平台 部署(通过集装箱)和长期可维护性(通过广泛的 文档)。此外，每个模块都将部署为独立的微服务 用于模块间通信的通用应用程序编程接口(API)格式和 与现有NCATS模块集成，包括图形用户界面。这些努力将 通过麻省理工学院和NCATS之间的密切合作来实现，以提高整体 NCATS ASPIRE平台的功能。

PROJECT SUMMARY Access to complex chemical matter (e.g., small molecule drug candidates) is a core requirement for testing biological hypotheses and probing human health. Current approaches to chemical synthesis rely on time-consuming planning and labor-intensive manual synthesis, which is a rate-limiting step in the discovery of new functional molecules. This collaborative project comprises the development of several virtual modules to support the multi-step chemical synthesis of new molecules in autonomous laboratories. These modules are designed to benefit traditional synthetic chemists in addition to automation chemists using the integrated hardware platform being developed by the ASPIRE team at NCATS. Computer-aided synthesis planning can be viewed as a hierarchical process of elaboration starting from the list of molecules of interest: (1) retrosynthetic planning to identify suitable starting materials and intermediates, (2) reaction condition recommendation to identify the conditions with which each reaction step should be run, (3) translation of hypothetical reaction steps into action sequences executable on automated hardware. Optional but valuable components include (4) recording procedures through an experimental planning module, (5) optimization of the timing and order of action sequences to most efficiently synthesize multiple synthetic targets via a digital twin of the platform, and (6) the iterative optimization of process parameters based on experimental responses in a feedback loop. This program will address each of these needs through the development of new software solutions employing state of the art algorithms in graph network theory, cheminformatics, deep learning for chemistry, and optimization. Software modules will be written using established software development best practices for ease of cross-platform deployment (via containerization) and long-term maintainability (via extensive documentation). Further, each module will be deployed as an independent microservice with a common application programming interface (API) format for inter-module communication and integration with existing NCATS modules, including graphical user interfaces. These efforts will be accomplished through close partnership between MIT and NCATS to enhance the overall capabilities of the NCATS ASPIRE platform.