Identifying multi-component drug candidates in natural products via association rule mining

OBJECTIVE To identify compound combinations as candidate multi-component drugs for the type 2 diabetes from natural product information.METHODS Chemical composition information of herbs in natural medicine was acquired by integrating conventional databases;Traditional Chinese Medicine Information Database(TCM-ID)and Traditional Chinese Medicine Integrated Database(TCMID).Therapeutic effect of each herb on the type 2 diabetes was examined by analyzing annotated function information with a text-mining method.The Apriori algorithm,which is a classical method for extracting associations between object in large-scale databases,was employed to infer association rules between compound combinations and therapeutic effect on the target disease.The chemical composition and therapeutic information of each herb was used as a transaction,which consists of the chemical compound combination as an antecedent item set and the therapeutic effect as a consequent item.The association rules with high support and confidence value were suggested as candidate multi-component drugs for the type 2 diabetes.RESULTS Totally 40 941 association rules were inferred with support lower bound 0.05% and maximum rule length 4.With respect to support and confidence,the top-ranked compound combination was puerarin and daidzin(support=0.15%,confidence=100%).In addition,the top 16 compound combinations were composed of 11 individual chemical compounds;puerarin,daidzin,abscisic acid,batatisine,dopamine,cholesterol,daidzein,gamma-aminobutyric acid,stigmasterol,campesteryl ferulate,and campesterol.To validate therapeutic effect of the proposed compound combinations,literature evidences of each individual compound were investigated.Among the 11 individual compounds,six compounds were reported to be effective for the treatment of the diabetes mellitus.CONCLUSION By analyzing natural product in formation with association rule mining,16 compound combinations are suggested as candidate multi-component drugs for the type 2 diabetes.These compound combinations are recommended for further investigation in the context of drug development.

Integrative database for multi-compound drug discovery in complementary medicine

OBJECTIVE To construct an integrative database for multi-compound drug discovery.METHODS We designed and constructed a database system,which integrates traditional herbal medicine,functional food,and drug combination information.Our database consists of six entity tables,namely drug combinations,functional foods,prescriptions,herbs,compounds and phenotypes.We established strategies for data integration and entity resolution to facilitate heterogeneous information of multi-compound therapies.To standardize the data,instances of entity tables are mapped to international identifiers,and phenotype terms in narrative text are extracted by using the named entity recognition(NER)method.RESULTS The database integrates therapeutic information of traditional herbal medicine,functional foods and combination drugs which is acquired from Traditional Chinese Medicine Information Database(TCM-ID),Food and Drug Administration(FDA)and Drug Combination Database(DCDB).The herb information is mapped to NCBI taxonomy identifiers,and compound information is mapped to PubChem and ChEMBL identifiers for standardization.We also applied MetaMap,a tool for recognizing UMLS concepts from narrative text,to extract phenotype terms.The current version of the database contains 6 291 drug combinations,1 615 functional foods,20 091 prescriptions,8889herbs,227 636 compounds and 11 744 phenotypes.CONCLUSION Our database provides various therapeutic information of multi-compound therapies which serve as a fundamental resource for the polypharmacology research.