目的为解决中药药性描述的抽象、模糊导致难以准确把握其本质特性的问题,提出一种基于多层前馈神经网络(BP神经网络)的药向量训练(quantitative model of traditional Chinese medicine’s properties based on BP neural network,QM-BP...目的为解决中药药性描述的抽象、模糊导致难以准确把握其本质特性的问题,提出一种基于多层前馈神经网络(BP神经网络)的药向量训练(quantitative model of traditional Chinese medicine’s properties based on BP neural network,QM-BP)模型,实现中药药性的量化表示。方法首先对中药及其对应的功效进行整理,获得"中药-功效"样本对;其次,构建"中药-药向量-功效"3层结构的QM-BP模型,并利用中药的药性数据对模型进行初始化;最后,基于QM-BP模型使用"中药-功效"样本进行训练,得到BP药向量。结果将《中药学》教材所涉及的474味中药及其528个功效基于QM-BP模型训练并结合临床分析,发现训练后得到的BP药向量比药性的初始量化值更能反映中药的属性特征。此外,由于BP药向量与词向量具有相似的性质,发现功效相似的药物对应的BP药向量在欧几里得距离中距离较近,而功效差异较大的中药药向量在欧几里得距离中距离较远。结论利用BP神经网络构建药向量训练模型,在中药药性与功效具有关联性的基础上,对药性量化值进行修正,以期使药性量化值更精确。今后可优化QM-BP模型并开展药对、复方分析,以期探明中药药性及组方配伍中蕴藏的内在规律。展开更多
The similarities of the non-linear chemical （NLC） fingerprints of Radix Glycyrrhizaes from four producing areas and eight other traditional Chinese medicines （TCMs） were calculated, using a systemic similarity cal...The similarities of the non-linear chemical （NLC） fingerprints of Radix Glycyrrhizaes from four producing areas and eight other traditional Chinese medicines （TCMs） were calculated, using a systemic similarity calculation method proposed and three other main calculation ones （Euclidean distance, correlation coefficient and included angle cosine）. All of the correlation coefficient similarities of different TCMs are higher than 0.952, and the included angle cosines are all higher than 0.962. So, both the conelation coefficient and included angle cosine similarities are unable to be used as the criteria for quantitatively evaluating the similarities of NLC fingerprints of TCMs. Although all of the Euclidean distance similarities of Berry Liquorices from four producing areas are less than 73, those of the other eight TCMs are all more than 180. The Euclidean distance cannot reflect the relative magnitudes of the feature differences in the NLC fingerprints very correctly. The systemic similarity method is the best among the four ones. All of the systemic similarities of Berry Liquorices from the four producing areas are higher than 0.962, while those of the other eight TCMs are all lower than 0.805, and the systemic similarity can reflect the differences between samples most faithfully, and can be used as a quantitative one evaluating the similarities of NLC fingerprints of TCMs, by which TCM could be distinguished and evaluated quickly, simply and exactly.展开更多
文摘目的为解决中药药性描述的抽象、模糊导致难以准确把握其本质特性的问题,提出一种基于多层前馈神经网络(BP神经网络)的药向量训练(quantitative model of traditional Chinese medicine’s properties based on BP neural network,QM-BP)模型,实现中药药性的量化表示。方法首先对中药及其对应的功效进行整理,获得"中药-功效"样本对;其次,构建"中药-药向量-功效"3层结构的QM-BP模型,并利用中药的药性数据对模型进行初始化;最后,基于QM-BP模型使用"中药-功效"样本进行训练,得到BP药向量。结果将《中药学》教材所涉及的474味中药及其528个功效基于QM-BP模型训练并结合临床分析,发现训练后得到的BP药向量比药性的初始量化值更能反映中药的属性特征。此外,由于BP药向量与词向量具有相似的性质,发现功效相似的药物对应的BP药向量在欧几里得距离中距离较近,而功效差异较大的中药药向量在欧几里得距离中距离较远。结论利用BP神经网络构建药向量训练模型,在中药药性与功效具有关联性的基础上,对药性量化值进行修正,以期使药性量化值更精确。今后可优化QM-BP模型并开展药对、复方分析,以期探明中药药性及组方配伍中蕴藏的内在规律。
基金Project(2009GJD20033) supported by the National Science and Technology Ministry of China
文摘The similarities of the non-linear chemical （NLC） fingerprints of Radix Glycyrrhizaes from four producing areas and eight other traditional Chinese medicines （TCMs） were calculated, using a systemic similarity calculation method proposed and three other main calculation ones （Euclidean distance, correlation coefficient and included angle cosine）. All of the correlation coefficient similarities of different TCMs are higher than 0.952, and the included angle cosines are all higher than 0.962. So, both the conelation coefficient and included angle cosine similarities are unable to be used as the criteria for quantitatively evaluating the similarities of NLC fingerprints of TCMs. Although all of the Euclidean distance similarities of Berry Liquorices from four producing areas are less than 73, those of the other eight TCMs are all more than 180. The Euclidean distance cannot reflect the relative magnitudes of the feature differences in the NLC fingerprints very correctly. The systemic similarity method is the best among the four ones. All of the systemic similarities of Berry Liquorices from the four producing areas are higher than 0.962, while those of the other eight TCMs are all lower than 0.805, and the systemic similarity can reflect the differences between samples most faithfully, and can be used as a quantitative one evaluating the similarities of NLC fingerprints of TCMs, by which TCM could be distinguished and evaluated quickly, simply and exactly.
