Multiscale ab initio QM/MM and Machine Learning Methods for Accelerated Free Energy Simulations

用于加速自由能模拟的多尺度从头 QM/MM 和机器学习方法

• 批准号: 10696727
• 负责人: Xintian Feng
• 金额: $ 67.8万
• 依托单位: Q-CHEM, INC.
• 依托单位国家: 美国
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-04-01 至 2025-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10696727
• 关键词: Acceleration; Biochemical Reaction; Biology; Chemicals; Chemistry; Complex; Computer software; Development; Discipline; Enzymes; Free Energy; Hour; Marketing; Methods; Modeling; Phase; Positioning Attribute; Process; Quantum Mechanics; Reaction; Surface; System; Time; Update; biological systems; cost; design; enzyme activity; improved; inhibitor; machine learning method; machine learning model; materials science; molecular modeling; mutant; programs; quantum; quantum chemistry; simulation; time interval; tool

Q-Chem is a state-of-the-art commercial computational quantum chemistry software program that has aided about 60,000 users in their modeling of molecular processes in a wide range of disciplines, including biology, chemistry, and materials science. In this proposal, we seek to significantly reduce the computational time (now around 500,000 CPU hours) required to obtain accurate free energy profiles of enzymatic reactions. Specifically, we propose to use a multiple time step (MTS) simulation method, where a low-level (and less accurate) quantum chemistry or machine learning model is used to propagate the system (i.e. move all atoms) at each time step (usually 0.5 or 1 fs), and then a high-level (i.e. more accurate and expensive) quantum chemistry method is used to correct the force on the atoms at longer time intervals. In this way, the simulation can be performed at the high- level energy surface in a fraction of time, compared with simulations performed only using the high-level quantum chemical method. In the Phase I proposal, we successfully re-parameterized low-level quantum chemistry models and developed machine learning models for MTS simulations. Through these developments, we were able to extend the high-level force update to only once every 8 fs or longer. In the Phase II period, we will further improve and automate the workflow for developing the low-cost models, which will further enhance the computational efficiency of our MTS simulations. In addition, these advances will be combined by the EnzyDock method to facilitate the study of multi-step enzyme reactions and the design of covalent/noncovalent inhibitors and mutant enzymes. The addition of these new tools will also further strengthen Q-Chem's position as a global leader in the molecular modeling software market, making our program the most efficient and reliable computational quantum chemistry package for simulating large, complex chemical/biological systems.

Q-Chem是一个最先进的商业计算量子化学 该软件程序已帮助约60，000名用户进行建模， 分子过程在广泛的学科，包括生物学，化学， 和材料科学。 在这个建议中，我们寻求显着减少计算时间（现在 大约500，000个CPU小时），以获得准确的自由能分布 酶促反应具体来说，我们建议使用多时间步长（MTS） 模拟方法，其中低水平（且不太准确）的量子化学或 机器学习模型用于传播系统（即移动所有原子）， 每个时间步长（通常为0.5或1 fs），然后是高级别（即更准确， 昂贵）量子化学方法用于校正原子上的力 以更长的时间间隔。通过这种方式，可以在高- 水平能量表面在一小部分时间，与模拟进行比较， 只使用高级量子化学方法。 在第一阶段的提案中，我们成功地重新参数化了低水平的量子 化学模型和开发的MTS模拟机器学习模型。 通过这些事态发展，我们得以延长高级别部队最新情况通报时间， 到每8 fs或更长时间仅一次。在第二阶段，我们将进一步提高 并使开发低成本模型的工作流程自动化，这将进一步 提高MTS模拟的计算效率。另外这些 这些进展将通过酶码头方法结合起来，以促进对 多步酶反应和共价/非共价抑制剂的设计 和突变酶。 这些新工具的加入也将进一步加强Q-Chem的地位， 分子建模软件市场的全球领导者，使我们的程序 最有效和可靠的计算量子化学包， 模拟大型复杂的化学/生物系统。