Phytochemical informatics of traditional Chinese medicine and therapeutic relevance.

Distribution patterns of 8411 compounds from 240 Chinese herbs were analyzed in relation to the herbal categories of traditional Chinese medicine (TCM), using Random Forest (RF) and self-organizing maps (SOM). RF was used first to construct TCM profiles of individual compounds, which describe their affinities for 28 major herbal categories, while simultaneously minimizing the level of noise associated with the complex array of diverse phytochemicals found in herbs from each category. Profiles were then reduced and visualized with SOM. The distribution of 10 major phytochemical classes, in relation to TCM profile, was delineated with SOM-Ward clustering. These classes comprised aliphatics, alkaloids, simple phenolics, lignans, quinones, polyphenols (flavonoids and tannins), and mono-, sesqui-, di-, and triterpenes (including sterols). Highly distinctive patterns of association between phytochemical class and TCM profile were revealed, suggesting that a strong phytochemical basis underlies the traditional language of Chinese medicine. Maps trained after random permutation of herbs assigned to each category were, by contrast, devoid of feature, providing additional evidence for the significance of these associations. Most classes were split into relatively few clusters, and further analysis revealed that simple descriptors, comprising skeletal type, molecular weight, and calculated log P, were in most cases able to readily discriminate within-class clusters. Relationships between TCM profile and predicted activities, relating to therapeutically important molecular targets, were explored and indicate that ethnopharmacological data could play an important role in pharmaceutical prospecting from Chinese herbs as well as identifying links between Chinese and Western medicine.