COMPUTER-ASSISTED STRUCTURE ELUCIDATION

生物分子的化学结构表征 和光谱特性在生物医学中具有广泛的重要性 以理工科为重 由于其复杂性，结构表征是全面的 这是一项耗时的活动，主要留给了 领域 这项研究的目标是大幅增加 这些专家的生产力。 鉴于专家的熟练程度，在容易挑选出正确的结构 一个未知的从有限数量的替代结构， 如果专家能够更迅速地指定结构， 可以开始的结构表征问题，这样一个概念上的 可管理的合理替代方案的数量。 因此， 研究是创造基于人工智能的计算机软件 能够直接降低集体光谱特性（红外， 质子和碳-13核磁共振，紫外和质量 光谱数据）的未知到概念上可管理的数量， 立体化学定义的结构，其中每一个都与所述立体化学定义的结构相容。 观察光谱数据。 专家仍然是一个必要的球员，但 省去了“手册”中耗时的部分 过程，即产生有限数量的替代品。 要创建的软件的主要功能包括：（1） 光谱解释和结构生成，（2）接受规定 专家已知的任何结构信息，以及（3） 处理现实世界的结构表征问题，而不考虑 未知数所属的结构类。 最后命名的要素 需要不依赖于光谱库的系统。

The characterization of the structure of a biomolecule on the its chemical and spectral properties is of widespread importance in the biomedical sciences. Because of its complexity, structure characterization is overall a time-consuming activity that has been largely left to experts in the field. The goal of this research is to substantially increase the productivity of those experts. Given the expert's proficiency in readily picking out the correct structure of an unknown from among a limited number of alternative structures, a much more expeditious assignment of structure would be achieved if the expert could begin the structure characterization problem with such a conceptually manageable number of plausible alternatives. Thus, the goal of the research is to create artificial intelligence based computer software capable of directly reducing the collective spectral properties (infrared, proton and carbon-13 nuclear magnetic resonance, ultraviolet and mass spectral data) of the unknown to a conceptually manageable number of stereochemically defined structures, each of which is compatible with the observed spectral data. The expert is still a required player, but is spared what is actually the time-intensive component of the "manual" process, that of producing the limited number of alternatives. Key features of the software to be created include, (1) programs for both spectrum interpretation and structure generation, (2) provision to accept any structural information known to the expert and (3) the capability to treat real-world structure characterization problems without regard to the class of structures to which the unknown belongs. The last named feature requires a system that is not spectral library dependent.