Mining Small Molecule Information For Drug Repurposing And Prediction Of Cancer-Related Proteins

Currently many open databases have accumulated large numbers of data on small molecules, including drug-drug interaction data and structural data of small molecules. Therefore, effective approaches are needed in order to mine these data for useful information. This study proposed two methods for mining small molecule information and applied them in the field of medicine. The first method used drug-drug interaction data for drug repurposing, and the second employed structural data of small molecules to predict the proteins relevant to cancer based on chemoinformatics.1. Drug repurposing based on drug-drug interaction. Given the high risk and lengthy procedure of traditional drug development, drug repurposing (finding new uses for old drugs) is gaining more and more attention. Although many types of drug information have been used to repurpose drugs, drug-drug interaction data, which imply possible physiological effects or targets of drugs, remain unexploited. In this work, similarity of drug interaction was employed to infer similarity of the physiological effects or targets for the drugs. We collected10835drug-drug interactions concerning1074drugs, and for700of them, drug similarity scores based on drug interaction profiles were computed and rendered using a drug association network with589nodes (drugs) and2375edges (drug similarity scores). The589drugs were clustered into98groups with Markov Clustering Algorithm, most of which were significantly correlated with certain drug functions. This indicates that the network can be used to infer the physiological effects of drugs. Furthermore, we evaluated the ability of this drug association network to predict drug targets. The results show that the method is effective for317of561drugs that have known protein targets. Comparison of this method with the structure-based approach shows that they are complementary. In summary, this study demonstrates the feasibility of drug repurposing based on drug-drug interaction data.2. Predicting the proteins relevant to cancer based on chemoinformatics. Cancer poses a great threat to us, but traditional treatments for cancer usually cause serious side effects. Therefore, it is very urgent to develop novel anticancer drugs that specifically target the tumor. This requires a comprehensive understanding of the mechanisms of cancer progression, so the identification of cancer-related proteins carries much significance. This study attempted to infer the proteins playing roles in cancer from a new perspective. In this research, chemoinformatics was employed to analyze the correlation between anticancer compounds and the active chemicals on separate proteins in open bioactivity databases, which might reveal the association between cancer and these proteins. After examining the list of predicted cancer-relevant proteins, it was found that this chemoinformatics approach could be well applied to the inference of important proteins in cancer. According to relevant reports, about a half of the31proteins with the best association values are associated with the proliferation, apoptosis, or differentiation of cancer cells. In addition, a chemical-protein matrix emerged during the prediction process, which consisted of all analyzed proteins and the3160active chemicals against K-562cancer cell line. The matrix contains the information on the possible interactions between proteins and chemicals, and could assist in drug discovery. To sum up, chemoinformatics can help to reveal cancer-related proteins and thus the molecular mechanisms underlying cancer.