choderalab/assaytools

Chembl prior centered around 0?

sonyahanson opened this issue · 5 comments

sonyahanson commented

Is the chembl prior centered around 0? If so, this seems wrong and we should adjust accordingly if we ever plan to use this: https://github.com/choderalab/assaytools/blob/master/AssayTools/pymcmodels.py#L372

Currently we generally use the uniform prior, so this hasn't been an issue thus far...

jchodera commented

Is the chembl prior centered around 0?

Nope. It's a fit to the ChEMBL distribution of assay data.

sonyahanson commented

Sorry, could you explain a bit more how a fit to ChEMBL data happens from this line of code?

DeltaG = pymc.Normal('DeltaG', mu=0, tau=1./(12.5**2))
jchodera commented

This is not really meant for production, so we should deprecate it.

I simply took the distribution of affinity data in ChEMBL from one of Christian Kaiser's papers (the same one in Figure 6 of my review on entropy-enthalpy compensation) and fit a Gaussian to it. Those are the parameters of the Gaussian.

jchodera commented

Hm, but they should be centered closer to -10 kcal/mol. So let's ditch this prior.

sonyahanson commented

Could we just change the mu to be closer to that of the meta-analysis? I'd be interested to see how a different prior effects our results.