minzhe / digresyn Goto Github PK

Hello,

I am having the problem of reproducing/validating the results from Nature biotechnology 32 (2014): 1213-1222. Could you, please, help me out of my difficulties?

According to the paper this method got the PC-index of 0.61303 (1st place in the challenge) and 0.59981 (3rd place). I tried various configurations and my PC-index falls in the range between ~0.43 to ~0.48, which means that the method performs (in my hands) a bit discordantly to the gold standard data or, at best, closely to the 50/50 prediction.

• I used both the demo data geneExp.demo and those downloaded from the DREAM7 challenge website.

• I used both doseRes.demo and the f2B values I interpolated myself from the images of the dose-response curves reported in the paper's Supporting Material (because actual dose-response point-by-point curves were not provided in the paper).

• I used both Media and DMSO as the negative control.

• I do log2-transform of the data.

• I tried fold change thresholds from 0.1 to 5.0.

• I used all time points from the data (6h, 12h, and 24h)

• I used my own PC-index calculator (I can give you the code, but the algorithm is pretty straightforward though and I do not expect any problems here).

Do you have any ideas on how to reproduce the results?

And, in any case, could you, please, provide for the exact data that were used to get the paper's highest PC-index (0.61303)?

Thank you in advance!