Drug set enrichment analysis

Question

[kakukeshi]

Hi,

I have a list of drugs ranked by a score ( from -1 to 1, higher is more correlated) defining the likelihood to target a specific disease.

If I have a list of drugs that I know that are used to treat this disease. Is there any way to determine the significance of the method to prioritize these drugs.

ranked list <- a,b,c,d,e,f,g,h,i...(n drugs)

known drugs <- b,i

Cheers

[Kevin Blighe]

You can try something like cogena, written by a colleague of mine. In the vignette, take a look at section 6: Drug repositioning

Kevin

[kakukeshi]

Thanks Kevin. Unfortunatelly, I don't understand really how can I apply this method to my problem. Can you explain a little more please

[Kevin Blighe]

Slightly misread your question and assumed that you had genes and wanted to see in which drug pathways they were appearing. If you currently have correlation values, then you can also derive a p-value from the correlation test - why not try that? The function in R (looks like you're using R) is cor.test

Knowing the values to what, exactly, you are correlating these drugs would also help. If you have some numerical representation of disease activity, for example, then you could easily build a linear regression model and refine it via stepwise regression in order to build an 'official' cross-validated predictive model with AUC. Take a quick look here: A: Resources for gene signature creation

Answer 1

[Zhilong Jia]

Giving the predicted drugs list, you can analyze the function of the drugs to check the association between disease and significantly enriched functions of the drugs lists (such as via Drug-responded genes/targets-pathway/GO-disease; drug-phenotype/side effect/disease). Recommended tools like http://dsea.tigem.it/ or http://www.cuilab.cn/drugpattern. I believe there are more if you search using the title you typed "Drug set enrichment analysis", maybe as an answer.

Based on the information (known drugs <- b,i), it seems the method is qualitatively good as the known drugs for the disease appeared in the top among the predicted drug list.

If you'd like to evaluate the prediction method itself systemically, AUC/F-score as mentioned by my colleague, @Kevin, will probably be useful.

Answer 2

[ivivek_ngs]

I would list out DGIdb, probably if you have not come across. Take a look at it. The one of TIGEM is good as well but everything depends on how well the resources are updated. Opentargets should have also the functionality, if I am not wrong. Check it once or write their support.

Now significance method depends on table that you generate be it with enrichment score and p-value and then you can make some diagnostic plots or rank based on the enrichment score and significance.

Answer 3

[Rashedul Islam]

You can use GESA tool using preranked option. Details is here.