About the calculation of AUC in the example ipynb--proximity.ipynb

[wangze09]

Hi Emre,
I have just read your wonderful work "Network-based in silico drug efficacy screening. Nat. Commun. 7:10331" and try to reproduce your codes.
I see you calculate the AUC in proximity.ipynb with list of -d$z and the list of ytrue in R script. I just wonder if the threshold of Z（-0.15）mentioned in the article should be used to judge the ypred(if z value <=-0.15 ypred=1，else ypred=0).
Thanks for your time!
Best wishes

Ze

[emreg00]

Hi Ze,

Yes, that was the cutoff we used in the end based on the observation that the drug targets are not necessarily significantly close to the disease genes (thus a relaxed z-score cutoff where the two node sets are closer but not statistically significant). This is because drugs tend to have low target specificity and drug targets interact with many proteins in biological networks that are known to be small world (where one can reach to any other node in the networks in few steps).

Hope this helps.

Best wishes,

Emre

[wangze09]

Hi Emre,
Yeah, I get your ideas. That means if I want to filter potential drugs for a certain disease, I need to set a relaxed z-score cutoff because 'a drug should be sufficiently selective (that is, proximal to the disease) to have therapeutic effect but not necessarily exclusive (significantly proximal to the disease).'
Thanks for your wonderful explanation!
Best wishes,

Ze