A Computational Systems Biology Approach for Identifying Candidate Drugs for Repositioning for Cardiovascular Disease

Short Communication


We report an in silico method to screen for receptors or pathways that could be targeted to elicit beneficial transcriptional changes in a cellular model of a disease of interest. In our method, we integrate: (1) a dataset of transcriptome responses of a cell line to a panel of drugs; (2) two sets of genes for the disease; and (3) mappings between drugs and the receptors or pathways that they target. We carried out a gene set enrichment analysis (GSEA) test for each of the two gene sets against a list of genes ordered by fold-change in response to a drug in a relevant cell line (HL60), with the overall score for a drug being the difference of the two enrichment scores. Next, we applied GSEA for drug targets based on drugs that have been ranked by their differential enrichment scores. The method ranks drugs by the degree of anti-correlation of their gene-level transcriptional effects on the cell line with the genes in the disease gene sets. We applied the method to data from (1) CMap 2.0; (2) gene sets from two transcriptome profiling studies of atherosclerosis; and (3) a combined dataset of drug/target information. Our analysis recapitulated known targets related to CVD (e.g., PPARγ; HMG-CoA reductase, HDACs) and novel targets (e.g., amine oxidase A, δ-opioid receptor). We conclude that combining disease-associated gene sets, drug-transcriptome-responses datasets and drug-target annotations can potentially be useful as a screening tool for diseases that lack an accepted cellular model for in vitro screening.


Atherosclerosis Gene expression analysis Drug repositioning Bioinformatics 

1 Introduction

Due to functional pleiotropy of drug targets and poly-pharmacological drug-target mechanisms, many approved drugs likely have undiscovered therapeutic applications. In light of the significant cost-efficiencies of bringing an already approved drug to market under a new indication (i.e., drug repositioning; [1]), the recent availability of comprehensive transcriptome profiling data for the effects of drugs on cell lines [2] has spurred interest in computationally screening for new indications for existing drugs, i.e., in silico drug repositioning [3].

Cardiovascular diseases (CVD) and their main underlying pathology, the chronic inflammatory disease atherosclerosis, are together the leading cause of death. To the extent that current lipid-lowering drugs for CVD prevention are estimated to reduce CVD mortality by only 20 % [4], there is a need for new therapeutic approaches. Atherosclerosis is particularly attractive for a computational approach because there is not a natural in vitro cellular assay for drug-to-phenotype screening for this disease. A key cellular constituent in atherosclerotic plaque, the macrophage (an innate immune cell of the myeloid lineage), is both an enticing therapeutic target and has an analogous human myeloid cell line, HL60, that is a workhorse cell line in pharmacology related to hematopoiesis [5].

In this work, we investigated whether candidate atheroprotective or cardioprotective drugs can be identified by applying a rank-based statistical test (gene set enrichment analysis or GSEA; [6]) to measurements of drug-induced differential expression of atherosclerosis-related genes (“gene sets”) in a physiologically relevant human cell line. For the transcriptome profiling data on cell line drug responses, we used measurements from the Connectivity Map 2.0 (CMap2) database [2] for differential expression of 12,135 genes in HL60 cells that were treated with vehicle or one of 1229 drugs. We used CVD-related gene sets from two transcriptome studies of human tissues: a study comparing three types of atherosclerotic arteries (aorta, coronary, and carotid) with normal arteries [7] (Cagnin et al., 161 genes), and a study comparing unstable versus stable carotid plaques as determined by specific molecular markers [8] (Puig et al., 1271 genes). Thus, our analysis included four sets of genes; two for genes that are up- or downregulated in atherogenesis, and two for genes that are up- or downregulated as plaque becomes unstable. Using the gene sets and the CMap2 data, we screened for drugs that reduced HL60 expression of genes in the “upregulated” gene sets and increased expression of genes in the “downregulated” gene sets, as measured by enrichment scores. We used a novel permutation-based approach to assess the significance of each enrichment score. We based our choice of weight factor for the analysis on accuracy results that we obtained by applying GSEA and a weighted Kolmogorov–Smirnov test with various weights to simulated data (the first such analysis of which we are aware).

2 Materials and Methods

2.1 Connectivity Map Analysis

We obtained probe intensity files (CEL files; Affymetrix HG-U133A and HT_HG-U133A GeneChips) for HL60 experiments spanning 1229 drugs (1406 files in all) from the Connectivity Map 2.0 website (broadinstitute.org/cmap). We mapped probe intensities to 12,135 probesets using the Entrez Gene-based probesets from the University of Michigan Custom CDF project (brainarray.mbni.med.umich.edu) release 18.0.0, and we obtained probeset-level intensities using the justRMA function in the Bioconductor software package “affy.” For each comparison of a drug to vehicle, we selected only probesets for which the within-sample-group-average log2 intensity is at least six (the background hybridization signal level) in either the drug or vehicle sample group. For each probeset, we computed the average log2 ratio of the intensity in drug-treated to vehicle-treated HL60 cells. Then, for each drug, we ranked all above-background probesets (genes) based on their log2 ratios. All subsequent analyses were carried out in the R statistical computing environment.

2.2 Synthetic Dataset Analysis to Determine Optimal Weighting

For the synthetic dataset analysis, we generated 1000 “positive” gene set ranks and 1000 “control” gene set ranks (each of size 80 genes). For both types of gene sets, rank assignments were randomly sampled from {1, …, 1229} without replacement. Rank assignments were selected with uniform probability for the case of a “control” gene set, and rank assignments for “positive” gene sets were selected with a bias probability for rank j defined by \(p_{j} = j^{ - q} /\sum\nolimits_{{j^{{\prime }} }} {(j^\prime)^{{- q}} }\), where the values of q tested were 0.1, 0.2, 0.3, and 0.4. Expression ratios rj for the set of 8000 genes were obtained from the average log2 ratios at each rank, across all drugs. The enrichment score E was computed for each of the 1000 gene sets, and gene sets were ordered by the E scores, and the area under the sensitivity versus false positive error rate (i.e., area under the receiver operating characteristic) curve was computed using the ROCR software package.

2.3 Selection of Disease-Associated Gene Sets

We obtained genes from Table S2 of the Cagnin et al. article [7] that were classified as “up” (64 genes) or “down” (97 genes), for the gene sets for coronary artery disease versus normal arteries. From Table S3 of the Puig et al. study [8], we selected genes that were classified as “inflamed” (900 genes) for the “up” gene set, and genes that were classified as “stable” (371 genes) for the “down” gene set. In both cases, genes were selected based on the differential expression analysis in the original study (FDR < 0.05 for the Cagnin et al. study and FDR < 0.12 for the Puig et al. study). The HGNC identifiers from the Cagnin et al. and the Puig et al. studies were converted to Entrez Gene identifiers using the web-based tool DAVID (david.ncifcrf.gov).

2.4 Permutation Method for Computing P values

For each drug and gene set resulting in a E score, we computed E (random) scores for 1000 randomly generated gene sets. For each random gene set, the genes’ ranks were sampled (uniform probability, no replacement). We computed the empirical P value of E as the cCDF of E in the distribution of E (random), using kernel density estimation (kCDF function in the sROC R package). For each pair of gene sets, a P value cutoff was determined at which the estimated false discovery rate would be 0.05 [9]. The resulting P value cutoffs for the gene set pairs based on the Cagnin et al. [7] and the Puig et al. [8] gene sets were P = 0.017 and P = 0.03, respectively.

2.5 TTD and DrugBank Databases

Drug-target data matrices were downloaded from two databases, DrugBank and Therapeutic Target Database (TTD), and matched to the drugs from the Connectivity Map. Drugs were matched using two different strategies, by its given name and by its CAS number. CAS number for each drug from the Connectivity Map was collected from DrugBank, Sigma-Aldrich and ChemSpider websites. Overall, 525 of 1087 drugs matched to DrugBank’s database and 568 of 1087 matched to TTD’s database. When combined, 604 of 1087 drugs were mapped to at least one target. Each drug target that was associated with at least three drugs was tested against the ranked list of drugs using GSEA.

3 Results and Discussion

Our method (Fig. 1) entails generating an overall disease association score ∆E for a given drug, based on two disease-related gene sets Sup and Sdown and based on the expression log2-ratios \(r_{i}\) of N genes (\(i \in \{ 1, \ldots ,N\}\)) in drug-treated versus vehicle-treated cells, where \(g_{j} \in \{ 1, \ldots ,N\}\) is the gene whose expression log2 ratio \(r_{{g_{j} }}\) has rank j (rank 1 means highest positive log2 ratio), as described below. For each gene set S (where S is either the set Sup for genes that are upregulated in disease versus normal tissue, or the gene set Sdown for genes that are downregulated in disease), we compute a weighted enrichment statistic E(S) by computing, for all \(i \in \{ 1, \ldots ,N = |S|\}\),
$$P_{i}^{\text{miss}} \,(S) = \sum\limits_{{\mathop {g_{j} \notin S}\limits_{j \le i} }} {\frac{{|r_{j} |^{zw} }}{{N_{M} }}} ; \quad P_{i}^{\text{hit}} \,(S) = \sum\limits_{{\mathop {g_{j} \in S}\limits_{j \le i} }} {\frac{{|r_{j} |^{w} }}{{N_{R} }}}$$
where \(w \ge 0\) is the weight factor for the expression ratio, the case z = 0 corresponds to the GSEA method, the case z = 1 corresponds to a weighted Kolmogorov–Smirnov test, and where the normalization factors \(N_{M}\) and \(N_{R }\) are defined by
$$N_{M} = \sum\limits_{{g_{j} \notin S}} {|r_{j} |}^{zw} ;\quad N_{R} = \sum\limits_{{g_{j} \in S}} {|r_{j} |}^{w} .$$
We compute the enrichment score E(S) by the maximum deviation from zero,
$$E\,(S) = {\text{extreme}}_{i} \left( {P_{i}^{\text{hit}} \,(S) - P_{i}^{\text{miss}} \,(S)} \right).$$
The overall drug-to-disease association score is the difference in the enrichment score between the two disease-associated gene sets,
$$\Delta E = E\,(S^{\text{up}} ) - E(S^{\text{down}} ).$$
Based on uncertainty in the literature regarding the optimal selection of the per-gene weight factor w for the statistical test (where w = 0 corresponds to the unweighted test) and on whether the unweighted test (z = 0) is more accurate than a weighted Kolmogorov–Smirnov (WKS) or weighted GSEA test (z = 1) [6], we simulated drug-response data and gene-set rankings under the null hypothesis (i.e., that the gene set is unrelated to the drug response) and under the alternate hypothesis (see Materials & Methods). We found that the unweighted GSEA (i.e., z = 0, w = 0) test was the most accurate (P < 10−4 for all pairwise comparisons with unweighted GSEA; Table 1), and thus we chose w = 0 and z = 0 as the parameter values.
Fig. 1

How we score a drug’s gene expression response against a disease-associated gene set. a Flow-chart of the overall analysis workflow. b Example enrichment score analysis test of gene expression responses of HL60 cells to the drug oxolinic acid when tested with a coronary artery disease gene set. c Rank positions of coronary artery disease genes among all genes, ranked by differential expression in response to the drug oxolinic acid in HL60 cells. Line plot, kernel density of the gene ranks. d Grayscale intensity scale bar for the −log2 gene expression ratios for oxolinic acid-treated versus vehicle-treated HL60 cells

Table 1

Accuracy results for detecting drug–disease associations within a synthetic dataset, for weighted Kolmogorov–Smirnov (K–S) and GSEA tests

Test type




AUC s.d.

Weighted KS





Weighted KS





Weighted KS

























Column headers are as follows: “Weighted KS,” weighted K–S test; AUC, area under the curve for sensitivity versus false positive error rate; AUC s.d., standard deviation of the AUC. The difference between the mean AUCs for the unweighted and GSEA (w = 0.25) test is significant at P < 10−4 (paired Student’s t test). The comparison of the unweighted and weighted w = 0.25 is also significant (P < 10−4). Shown here are the results for the rank bias parameter value q = 0.2 (see Materials and Methods), but the relative accuracy results using all other values of q were consistent with the above results

To rank drugs by their potential therapeutic benefit in the context of a specific pathophysiological process (SPP) (atherogenesis for the Cagnin et al. gene sets; plaque destabilization for the Puig et al. gene sets) [7, 8], we used a drug’s ΔE score and P value. Specifically, for each SPP, we ranked drugs by their ΔE values, such that a top-ranked drug would have effects on HL60 cells that are the most anti-correlated with the “directions” of the gene sets for the SPP. Additionally, for each drug and gene set, we tested the null hypothesis that there is no association between the gene set and the drug response by computing E for the gene set for randomly permuted gene ranks, from which we empirically determined a P value. We eliminated any drug whose P values for both of the atherogenesis [7] gene sets did not satisfy P < 0.017 or whose P values for both of the destabilization [8] sets did not satisfy P < 0.03 (reflecting the latter SPP’s larger gene set size). For each remaining drug, we averaged its ranks for the two SPPs (atherogenesis and plaque destabilization) to obtain an overall drug ranking; the top ten drugs are shown in Table 2 (the complete set of quantitative results for all drugs is provided as Online Resource 1).
Table 2

Ten top-ranked drugs based on our drug repositioning screen using measured drug responses in HL60 cells

Drug name

CAS identifier

Cagnin ∆E score

Cagnin up P value

Cagnin down P value

Puig up P value

Puig down P value

Puig ∆E score

Cagnin rank

Puig rank

Average rank

Trichostatin A (*)



5.66 × 10−17

4.58 × 10−35

5.82 × 10−22

7.08 × 10−66





Valproic acid (*)



1.34 × 10−8

9.66 × 10−5

5.08 × 10−61

1.60 × 10−45








1.14 × 10−3

1.29 × 10−11

6.15 × 10−37

1.08 × 10−42








5.81 × 10−10

2.62 × 10−79

5.37 × 10−76

2.41 × 10−3





Dehydrocholic acid



6.97 × 10−10

1.77 × 10−4

1.54 × 10−41

7.91 × 10−13





Trichostatin A (*)



8.73 × 10−4

1.74 × 10−8

1.60 × 10−3

2.13 × 10−50








1.58 × 10−2

2.43 × 10−52

9.06 × 10−17

3.22 × 10−5





Trolox C



3.05 × 10−3

3.65 × 10−5

1.91 × 10−15

1.40 × 10−33








1.93 × 10−9

3.79 × 10−3

7.76 × 10−131

8.01 × 10−14








6.58 × 10−4

1.31 × 10−80

1.61 × 10−69

1.16 × 10−2





Drugs’ ∆E score ranks were separately computed for the two SPPs and then averaged. HDAC inhibitors (denoted by *) scored well in the rankings. The lowest rank corresponds to the highest significance. For trichostatin A, ‡ = drug from Sigma; † = drug from CalBiochem

Based on the observation that the two top-scoring drugs share a common mechanism (histone deacetylase inhibition), we investigated whether there are other drug-target receptors or pathways that are enriched among the top-scoring drugs. For each drug, we obtained its list of annotated target receptors and pathways, based on information from two databases, DrugBank [10] and TTD [11]. We then applied the unweighted GSEA method to the ranks of drugs that are associated with a specific target, versus the ranks of the complete set of drugs, to obtain an E score for the drug target (i.e., the receptor or pathway). The top-scoring drug targets (Table 3) include PPARγ (whose pharmacological activation has been shown to reduce heart attacks [12]) and HMG-CoA reductase (the rate-limiting enzyme in cholesterol biosynthesis, and the primary target of the statin class of drugs for CVD prevention and treatment; this finding is also consistent with reports that statins exert anti-inflammatory effects in macrophages [13]). Other specific molecular targets, like amine oxidase [14], cAMP and cAMP-inhibited cGMP 3′,5′-cyclic phosphodiesterase 10A [15], and phospholipase A2 [16], are also reported to have roles in atherosclerosis.
Table 3

High-scoring drug targets, based on an unweighted GSEA test of a drug target against the complete list of drugs ranked by their E scores for CVD (for E > 0.5)


E score

# of drugs

Amine oxidase [flavin-containing] A



Peroxisome proliferator-activated receptor gamma



3-Hydroxy-3-methylglutaryl-coenzyme A reductase



cAMP and cAMP-inhibited cGMP 3′,5′-cyclic phosphodiesterase 10A



Phospholipase A2






δ-Opioid receptor



4 Conclusions

Our finding that two of the top-scoring drugs (ranked by their transcriptional effects in HL60 cells) are HDAC inhibitors is intriguing in light of the broad range of pathophysiological processes that involve HDACs in CVD [17, 18]. Of relevance to atherosclerosis, in macrophages, regulation of cholesterol metabolic enzymes in response to statins is mediated by HDAC repression that can be recapitulated by trichostatin A treatment [19]. More generally, HDACs have been reported to have anti-inflammatory effects on macrophages [20]. The other HDAC inhibitor, valproic acid, has recently been demonstrated to attenuate atherosclerosis in a mouse model [21]. While in this example application of our approach, we analyzed the effects of drugs on expression levels of protein-coding genes, our approach could in principle be readily applied to screen for drugs that modulate expression levels of disease-associated microRNAs (which are of significant interest in drug discovery for cancer and other diseases [22, 23, 24] using data from a drug-microRNA expression database such as Pharmaco-miR [25].

Our analysis shows that our computational approach, including the novel permutation test that we have introduced, is both practical and effective for identifying novel potential drugs and drug targets for two pathophysiological processes related to CVD (atherosclerosis and plaque destabilization). Our work also provides evidence in support of using unweighted GSEA for in silico drug screening, and our overall computational approach is readily applicable to other disease contexts.



This work was supported by the US National Institutes of Health (award HL098807 to S.A.R.), the Medical Research Foundation of Oregon (New Investigator Grant award to S.A.R.), Oregon State University (Division of Health Sciences Interdisciplinary Research Grant award to S.A.R. and University Honors College DeLoach Work Scholarship to A.Y.), the National Science Foundation (award numbers 1557605-DMS and 1553728-DBI to S.A.R.), and the Oregon State University Center for Genome Research and Biocomputing.

Supplementary material

12539_2016_194_MOESM1_ESM.pdf (341 kb)
Online Resource 1Quantitative results from an analysis of transcriptional responses of HL60 cells to 1229 drugs in which drugs were scored based on enrichments of genes from two atherosclerosis-related gene sets (Cagnin et al. and Puig et al. gene sets [7, 8]) among genes ordered by differential expression in response to drug treatment. P values are given in −log10 scale. (PDF 341 kb)


  1. 1.
    Yarchoan M, Arnold S (2014) Repurposing diabetes drugs for brain insulin resistance in Alzheimer disease. Diabetes 63:2253–2261CrossRefPubMedPubMedCentralGoogle Scholar
  2. 2.
    Lamb J, Crawford ED, Peck D, Modell JW, Blat IC et al (2006) The connectivity map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313:1929–1935CrossRefPubMedGoogle Scholar
  3. 3.
    Li J, Zheng S, Chen B, Butte AJ, Swamidass SJ, Lu Z (2015) A survey of current trends in computational drug repurposing. Brief Bioinform (ePub ahead of print)Google Scholar
  4. 4.
    Kostis WJ, Cheng JQ, Dobrzynski JM, Cabrera J, Kostis JB (2012) J Am Coll Cardiol 59(6):572–582CrossRefPubMedGoogle Scholar
  5. 5.
    Sun H, Wang Y (1995) Apoptosis of human leukemic HL-60 cells induced to differentiate by treatment with RA or DMSO. Cell Res 5:181–186CrossRefGoogle Scholar
  6. 6.
    Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL et al (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA 102:15545–15550CrossRefPubMedPubMedCentralGoogle Scholar
  7. 7.
    Cagnin S, Biscuola M, Patuzzo C, Trabetti E, Pasquali A et al (2009) Reconstruction and functional analysis of altered molecular pathways in human atherosclerotic arteries. BMC Genom 10:13CrossRefGoogle Scholar
  8. 8.
    Puig O, Yuan J, Stepaniants S, Zieba R, Zycband E et al (2011) A gene expression signature that classifies human atherosclerotic plaque by relative inflammation status. Circ Cardiovasc Genet 4:595–604CrossRefPubMedGoogle Scholar
  9. 9.
    Benjamini Y, Hochberg Y (1994) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B 57(1):289–300Google Scholar
  10. 10.
    Law V, Knox C, Djoumbou Y, Jewison T, Guo AC et al (2014) DrugBank 4.0: shedding new light on drug metabolism. Nucl Acids Res 42:1091–1097CrossRefGoogle Scholar
  11. 11.
    Zhu F, Shi Z, Qin C, Tao L, Liu X et al (2012) Therapeutic target database update 2012: a resource for facilitating target-oriented drug discovery. Nucleic Acids Res 40:1128–1136CrossRefGoogle Scholar
  12. 12.
    Dormandy JA, Charbonnel B, Eckland DJA, Erdmann E, Massi-Benedetti M, Moules IK et al (2005) Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study (PROspective pioglitAzone Clinical Trial In macroVascular Events): a randomised controlled trial. Lancet 366(9493):1279–1289CrossRefPubMedGoogle Scholar
  13. 13.
    Tuomisto TT, Lumivuori H, Kansanen E, Hakkinen SK, Turunen MP, van Thienen JV et al (2008) Simvastatin has an anti-inflammatory effect on macrophages via upregulation of an atheroprotective transcription factor, Kruppel-like factor 2. Cardiovasc Res 78(1):175–184CrossRefPubMedGoogle Scholar
  14. 14.
    Meszaros Z, Karadi I, Csanyi A, Szombathy T, Romics L, Magyar K (1999) Determination of human serum semicarbazide-sensitive amine oxidase activity: a possible clinical marker of atherosclerosis. Eur J Drug Metab Pharmacokinet 24:299–302CrossRefPubMedGoogle Scholar
  15. 15.
    Aizawa T, Wei H, Miano JM, Abe J, Berk BC, Yan C (2003) Role of phosphodiesterase 3 in NO/cGMP-mediated antiinflammatory effects in vascular smooth muscle cells. Circ Res 93:406–413CrossRefPubMedGoogle Scholar
  16. 16.
    Zalewski A, Macphee C (2005) Novel approaches to the treatment of dyslipidemia role of lipoprotein-associated phospholipase A2 in atherosclerosis biology, epidemiology, and possible therapeutic target. Arterioscler Thromb Vasc Biol 25(5):923–931CrossRefPubMedGoogle Scholar
  17. 17.
    Eom GH, Kook H (2014) Posttranslational modifications of histone deacetylases: implications for cardiovascular diseases. Pharmacol Ther 143(2):168–180CrossRefPubMedGoogle Scholar
  18. 18.
    McKinsey T (2012) Therapeutic potential for HDAC inhibitors in the heart. Annu Rev Pharmacol Toxicol 52:303–319CrossRefPubMedGoogle Scholar
  19. 19.
    Feig JE, Shang Y, Rotllan N, Vengrenyuk Y, Wu C, Shamir R et al (2011) Statins promote the regression of atherosclerosis via activation of the CCR7-dependent emigration pathway in macrophages. PLoS ONE 6(12):e28534CrossRefPubMedPubMedCentralGoogle Scholar
  20. 20.
    Wu C, Li A, Leng Y, Li Y, Kang J (2012) Histone deacetylase inhibition by sodium valproate regulates polarization of macrophage subsets. DNA Cell Biol 31:592–599CrossRefPubMedGoogle Scholar
  21. 21.
    Bowes AJ, Khan MI, Shi Y, Robertson L, Werstuck GH (2009) Valproate attenuates accelerated atherosclerosis in hyperglycemic apoE-deficient mice: evidence in support of a role for endoplasmic reticulum stress and glycogen synthase kinase-3 in lesion development and hepatic steatosis. Am J Pathol 174:330–342CrossRefPubMedPubMedCentralGoogle Scholar
  22. 22.
    Zeng X, Zhang X, Zou Q (2016) Integrative approaches for predicting microRNA function and prioritizing disease-related microRNA using biological interaction networks. Brief Bioinform 17:193–203CrossRefPubMedGoogle Scholar
  23. 23.
    Liu Y, Zeng X, He Z, Zou Q (2016) Inferring microRNA-disease associations by random walk on a heterogeneous network with multiple data sources. IEEE/ACM Trans Comput Biol Bioinform. doi:10.1109/TCBB.2016.2550432 Google Scholar
  24. 24.
    Wang Q, Wei L, Guan X et al (2014) Briefing in family characteristics of microRNAs and their applications in cancer research. Biochim Biophys Acta 1844:191–197CrossRefPubMedGoogle Scholar
  25. 25.
    Rukov JL, Wilentzik R, Jaffe I et al (2014) Pharmaco-miR: linking microRNAs and drug effects. Brief Bioinform 15:648–659. doi:10.1093/bib/bbs082 CrossRefPubMedGoogle Scholar

Copyright information

© International Association of Scientists in the Interdisciplinary Areas and Springer-Verlag Berlin Heidelberg 2016

Authors and Affiliations

  1. 1.Department of Biomedical SciencesOregon State UniversityCorvallisUSA
  2. 2.School of Electrical Engineering and Computer ScienceOregon State UniversityCorvallisUSA

Personalised recommendations