A Drug Repositioning Algorithm Based on Deep Auto-Encoder and Adaptive Fusion

Abstract

Background: Drug repositioning has aroused extensive attention by scholars at home and abroad due to its effective reduction in development cost and time of new drugs. However, the current drug repositioning based on computational analysis methods is still limited by the problems of data sparse and fusion methods, so we use autoencoders and adaptive fusion methods to calculate drug repositioning.Results: In this paper, a drug repositioning algorithm based on deep auto-encoder and adaptive fusion has been proposed against the problems of declined precision and low-efficiency multi-source data fusion caused by data sparseness. Specifically, the drug is repositioned through fusing drug-disease association, drug target protein, drug chemical structure and drug side effects. To begin with, drug feature data integrated by drug target protein and chemical structure were processed with dimension reduction via a deep auto-encoder, to obtain feature representation more densely and abstractly. On this basis, disease similarity was computed by the drug-disease association data, while drug similarity was calculated by drug feature and drug-side effect data. Besides, the predictions of drug-disease associations were calculated using a Top-k neighbor method that is more suitable for drug repositioning. Finally, a predicted matrix for drug-disease associations has been acquired upon fusing a wide variety of data via adaptive fusion. According to the experimental results, the proposed algorithm has higher precision and recall rate in comparison to DRCFFS, SLAMS and BADR algorithms that use the same data set for computation.Conclusion: our proposed algorithm contributes to studying novel uses of drugs, as can be seen from the case analysis of Alzheimer's disease. Therefore, it can provide a certain auxiliary effect for clinical trials of drug repositioning