Drug Repositioning Based On Systems Biology

Advances in science and technology have led to a growing understanding of diseases and drugs,but the speed at which these advances are transformed into the ability to treat diseases has been slower than expected.Drug research and development has been faced with the challenges of a high cost,high failure rate,and long development cycle.Drug repositioning,as a strategy to find new indications for known drugs,can significantly shorten the entire drug development cycle and reduce R&D costs and risks.Historically,drug repurposing has been largely opportunistic and serendipitous.In recent years,with the accumulation of biomedical data,drug repositioning based on computational methods has been able to provide meaningful guidance for drug development.This article mainly integrates multi-level heterogeneous data from the perspective of the network,and used the methods of systems biology to study drug repositioning.It mainly includes two parts:single drug repositioning and combination drug prediction.(1)Because the pathogenesis of diseases and drug mode of action(Mo A)have been revealed to be tightly connected with gene modules and there may be cross-talks among different functional modules,and current methods do not make full use of this information.So,we developed MNBDR(Module Network Based Drug Repositioning),a new method based on a module network,which combines the module network with the random walk algorithm,the important modules in disease occurrence and drug response are mined and applied to drug discovery.The results on different data sets show that the cross-talks between gene modules can indeed improve the prediction performance of the model.We then analyzed the robustness of MNBDR to further verify the reliability of our model.In addition,we also performed functional analysis on the important modules found by MNBDR,which indicated our method could reveal biological mechanisms in drug response.(2)The identification of effective drug combinations for specific diseases from many possible drug-drug-disease combinations is highly challenging.Here,we first propose a new feature(ie " friendship feature")to describe a drug,which integrates different types of drug data and similar information between the drug and other drugs.By quantifying the drug-drug relationship,we found that the disease has certain constraints on the single drugs that constitute its effective drug combination(we define the constraint as "drug composability"),and the effective combination of drugs will not be particularly similar.Based on this discovery,we propose a two-step strategy to identify clinically effective drug combinations for specific diseases.We first predict the single drugs that can be combined drugs and then calculate the combination of these drugs.At the same time,to avoid the bias caused by marking unknown samples as negative,we use OCSVM as the classification model.The results on hypertension and cancer datasets show that our strategy can significantly improve the prediction performance compared with other latest drug combination prediction methods.Finally,we further analyzed the potential drug combinations found by the model and found that the drug combinations predicted by the model can significantly enrich the known combination drugs and have lower toxicity,and it is consistent with the performance of the effective drug combination.In summary,for single drug repositioning and combination drug prediction,based on reasonable assumptions,we proposed a single drug screening method based on a module network and a two-step strategy based on friendship feature to solve these two problems and achieved excellent performance.Biological annotations further explain the reasons why the model has achieved better results,and provide new insights for the research of drug repositioning.