利用稀疏表达学习挖掘中医方剂功效配伍

Utilizingsparse representation learning to mine oriented efficacy compatibility in traditional chinese medicine prescriptions

摘要:

中医方剂是中医药学的重要组成部分，也是中医临床治病的主要形式和手段。为了“辨证论治”，需要从配伍功效出发，研究药组的配伍规则。多味药组成的方剂的功效不是其组成药物功效的简单叠加，而是由它们之间相互作用的结果。目前利用数据挖掘技术挖掘研究方剂的配伍，主要利用方剂中药物的频率，进行浅层分析，但这种方法并不能很好的揭示药物之间的相互联系。为此，本文提出了一种利用稀疏表达学习，自动挖掘古方中的功效配伍规律。稀疏表达学习结合L1正则化和逻辑斯蒂判别式，将不起作用或作用很小的药物视为是噪声过滤掉，起主导作用的药物则为被挖掘的功效配伍药组。最后，将提出的方法在14种功效的古方数据集中进行实验和验证，并以Dice系数和平均查准率作为评估参数，实验结果证明稀疏表达学习方法相比目前的主流方法在配伍规则的挖掘上更准确、有效。

Abstract:

Traditional Chinese medicine (TCM) prescription, as an important part of TCM theory, is one of the main manifestation forms and ways of clinical treatment. We need to study orientedefficacy compatibility for treatment based on syndrome differentiation. A prescription is composed of several or a dozen drugs, which efficacies are not simply the composition of all individual effect. In fact, its efficacies are the results of the interactions among drugs inside the prescription. At present, most researches focus on exploiting the frequencies of drugs in prescriptions by utilizing data mining technologies, which cannot catch the interactions among drugs. Therefore, this paper proposes a novel algorithm utilizing sparse representation learning to mine oriented efficacy compatibility in TCM ancient prescriptions, which takes low weight drugs as noise and makes up an oriented efficacy drug group with high weight drugs. We combine the logistics and L1 norm based regularization to mine the oriented efficacy compatibility. Lastly, 14 prescription datasets with different efficacies are used to validate our approach as well as dice index and the average retrieval rate are taken as metrics. Experimental results show that our approach is more effective and accurate than those of the stateoftheart research.