Drug Repositioning Based On Transcriptome Data

Although improvement in the sensitivity of computational drug repositioning methods has identified numerous credible repositioning opportunities,few have been progressed.Failure of progression can be explained by several reasons besides the common issue of limited scope for intellectual property.The unavailability of an integrative drug repositioning database largely prevents experimental researchers in drug development efficiently access and translate these results;Arguably the “black box” nature of drug action in a new indication based on transcriptome-based drug repositioning methods can be a barrier to the required investment for pre-clinical studies and clinical trials;The inefficiency of gene ranking methods is one of critical causes for producing the false positive results in drug repositioning.These issues highlight the urgent need for constructing an open drug repositioning database,identifying core target genes of disease,exploring drug mode of action in the disease context as well as effectively modelling the gene ranking question.I firstly proposed co-expressed gene set enrichment analysis(cogena)and released the cogena package in the Bioconductor community.By integrating co-expression information and several gene sets to perform the enrichment analysis,cogena can substantially improve our understanding of gene function in multiple perspectives.Firstly,cogena was applied to co-expressed gene pathway analysis,drug target discovery and target-based drug repositioning of Parkinson?s disease(PD).Accordingly,I obtained 39 core target genes for PD,among which 13 genes are the targets of approved drugs including neuroprotective drugs,such as quercetin,cholic acid and oprozomib.Secondly,I hypothesized that anti-correlation between expression signatures of the perturbated genes of a drug and the key co-expressed genes of a disease indicated the drug is likely to treat the disease,and then designed a drug repositioning pipeline using cogena,implying the drug mode of action.Applying this pipeline to psoriasis skin transcriptomics analysis,I computationally identified two approved psoriasis drugs with different modes of action,methotrexate and ciclosporin,predicted several drugs for psoriasis and explained their modes of action,such as cell cycle blockers resveratrol,ciclopirox,etoposide and trifluridine.Moreover,this pipeline was applied to methotrexate.Similar cell cycle blockers were obtained.Additionally,6 indications of this drug were recoveried and several new indications were predicted.Thirdly,I implemented a drug repositioning database and web server,D2 Dpage,by integrating 7275 gene expression signatures of drugs and diseases and calculating the similarity between them using Parametric Analysis of Gene Set Enrichment,PAGE.And I proved the utility and performance with several examples.For example,to discover drugs for psoriasis,I identified the biologics etanercept and off-label drug pimecrolimus.Therefore,D2 Dpage is the first drug repositioning database and web server including many gene expression signatures of drugs and diseases.Fourthly,based on the previous study on psoriasis,I integratively analyzed the PAGE-based correlations between co-expression genes of psoriasis,the gene expression signatures of biologics and autoimmune diseases,computationally identified the approved biologics etanercept and ixekizumab for psoriasis,evaluated the performance of unapproved biologics brodalumab and guselkumab for treating psoriasis,obtained methotrexate and off-label psoriasis drug isotretinoin and predicted several psoriasis drugs,consistent with the results of cogena analysis based on the psoriasis transcriptome data.Furthermore,I identified co-expressed genes robustly existing in the transcriptome of psoriasis,biologics and several autoimmune diseases,resulting in 32 core target genes of psoriasis.Additionally,I carried out targets-based drug repositioning and predicted several candidate drugs for psoriasis,including etoposide targeting TOP2 A.Finally,I applied Discriminant Non-negative Matrix Factorization,DNMF,into gene ranking with reasonably biological context.The proposed method outperformed other widely used methods in gene ranking.In addition,I released DNMF R package in the CRAN.