蛋白质紫外吸收光谱的机器学习模拟

蛋白质的结构决定其功能，因此“蛋白质结构测定”是现代科学的重要问题之一。通过发展快速响应的紫外吸收光谱技术，人们可以表征蛋白质的结构和动力学信息。然而蛋白质复杂多变的结构、实时动态的演化和环境涨落，使得蛋白质紫外吸收光谱的模拟存在巨大的计算瓶颈。申请人拟结合量子化学和机器学习方法，突破传统计算化学方法在蛋白质电子光谱领域的计算瓶颈问题，发展快速预测蛋白质紫外吸收光谱的方法。申请人首先通过分块方法将蛋白质拆分成肽键和氨基酸残基子片段，其次使用量子化学构建子片段具有化学精度的数据库并提取描述符，采用机器学习方法进行特征学习，继而基于激子哈密顿策略将子片段重新耦合，可以实现蛋白质紫外吸收光谱高效而准确地预测。新兴的人工智能技术与传统的量子化学方法的结合，可以充分利用新兴信息技术解决生命科学和生物化学的重大问题，谱写蛋白质光谱的崭新篇章。

"Protein structural determination" is considered as one of the most important issues in modern science because the functions of proteins are determined by their structures. We can characterize the structures and dynamics information of proteins by developing a real-time ultraviolet absorption spectroscopy technique. However, a huge computational bottleneck occurs when we calculate ultraviolet absorption spectroscopy for proteins due to their complex and flexible structures, real-time dynamic evolution and environmental fluctuations. This project will employ a combination of quantum chemistry and machine learning method to address the computational chemistry challenge of ultraviolet absorption spectra simulation in proteins domain and further develop an efficient way to calculate the molecular spectra of proteins. Proteins are firstly divided into peptides and amino acids residues fragments based on divided-and-conquer methods, and then a feature learning is carried out to establish structural-property relationship between molecular descriptors and the corresponding spectroscopy database calculated and extracted from quantum chemistry simulations of the divided fragments. Afterwards, the individual fragments can be recombined and coupled based on Frenkel exciton model to build the exciton Hamiltonian matrix for proteins and further generate an efficient and accurate framework for ultraviolet absorption spectroscopy simulations of proteins. We believe that our project can tackle some significant problems in life sciences and biochemistry by employing the novel strategy derived from the combination of emerging artificial intelligence technologies and traditional quantum chemistry methods, and further provide a potential method for the molecular spectroscopy simulation of proteins.