Data-driven modeling of the vibrational spectroscopy of ion channels

离子通道振动光谱的数据驱动建模 摘要 本研究计划的长期目标是（1）开发新的计算方法，以准确模拟， 的线性和二维红外（2D IR）光谱的蛋白质，并使用开发的方法， 模拟和设计新的二维红外实验（2）研究离子传输和分子机制， KcsA离子通道选择性的典型起源和（3）阐明了KcsA离子通道的构象和水化变化。 电压激活期间KvAP通道的电压感应域。尽管经过几十年的研究，我们仍然 没有关于离子通道动力学和施加电压对离子通道的影响的直接信息 结构.二维红外光谱是一种新兴的分析技术，它可以用化学方法探测蛋白质动力学， Cal键特异性空间和高时间分辨率。2D IR光谱类似于NMR光谱， 除了它使用红外光脉冲来测量振动，而不是用于核辐射的脉冲磁场。 旋转新方法的改进扩展了二维红外光谱学的前沿，允许研究com-IR光谱。 复杂的生物系统。特别令人感兴趣的是NMR和 X射线晶体学很难应用，例如离子通道。解释密集的2D IR光谱是困难的 没有模拟，可以定量连接光谱特征的原子结构模型。目前， 蛋白质的2D IR光谱使用模型驱动的，主要是经验的，光谱图进行建模， 溶剂诱导电场和骨架二面角对振动频率和耦合的影响。这个ap- 然而，PROACH缺乏系统的可改进性，具有有限的可转移性，最多提供定性的准确性， 并且对于异质环境中的肽是不准确的。从模型驱动的范式转移， 我们将使用基于图形神经网络的从头开始数据驱动方法来准确地建模 蛋白质在真实环境中的振动光谱。所提出的方法将提供计算支持， 端口正在进行和未来的离子通道的二维红外实验。拟议研究的结果将 显著增强我们对离子通道功能的分子水平机制的理解。大 由离子通道功能缺陷引起一系列神经、心血管和肌肉疾病。一 更好地了解这些疾病的起源将为改进靶向离子治疗铺平道路。 渠道

Data-driven modeling of the vibrational spectroscopy of ion channels Abstract The long-term goals of this research program are (1) to develop new computational methods for accurate sim- ulations of linear and two-dimensional infrared (2D IR) spectra of proteins and use the developed methods to simulate recent and design new 2D IR experiments to (2) investigate the mechanisms of ion transport and molec- ular origins of selectivity in the KcsA ion channel and (3) elucidate the conformational and hydrational changes of the voltage-sensing domain of the KvAP channel during voltage activation. Despite decades of research, we still don’t have the direct information on ion channel dynamics and the effects of an applied voltage on ion channel structures. 2D IR spectroscopy is an emerging analytical technique that probes protein dynamics with chemi- cal bond-specific spatial and high temporal resolution. 2D IR spectroscopy is analogous to NMR spectroscopy, except that it uses pulses of infrared light to measure vibrations rather than pulsed magnetic fields for nuclear spins. New methodology improvements expand the frontiers of 2D IR spectroscopy, permitting the study of com- plex biological systems in their native environments. Particularly interesting are systems for which NMR and X-ray crystallography are difficult to apply, such as ion channels. Interpreting congested 2D IR spectra is difficult without simulations that can quantitatively connect spectral features to atomistic structural models. Currently, 2D IR spectra of proteins are modeled using model-driven, mostly empirical, spectroscopic maps that correlate solvent-induced electric field and backbone dihedral angles to vibrational frequencies and couplings. This ap- proach, however, lacks systematic improvability, has limited transferability, provides qualitative accuracy at best, and is inaccurate for peptides in heterogeneous environments. Shifting away from the model-driven paradigm, we will use ab initio-based data-driven approaches based on Graph Neural Networks to accurately model the vibrational spectra of proteins in realistic environments. The proposed methods will provide computational sup- port for the ongoing and future 2D IR experiments on ion channels. The results of the proposed studies will significantly enhance our understanding of the molecular-level mechanisms of function of ion channels. A large spectrum of neurological, cardiovascular, and muscle disorders result from defective ion channel functioning. A better understanding of the origins of these diseases will pave the way for improved therapeutics that target ion channels.