Could we just set a slightly more restrictive prior on F_PL?
sonyahanson opened this issue · 1 comments
Tested the effect of setting F_PL
using values taken from the mean after equilibration from a previous run of assaytools, and dF_PL=0.50
, which was approximated from the standard deviation. Tested this for Src:Erlotinib and Src: Gefitinib.
F_PL_def = {'Src-Erlotinib-EF': 59601584903.0,
'Src-Gefitinib-GH': 81835891154.8}
Similar to issue #106 we see an expected removal of long tails and better reproducibility compared to without setting the F_PL
. Note here, we are still looking at a dPstated = 0.35 * inputs['Pstated']
.
Without setting F_PL
:
Setting F_PL
and dF_PL
:
Without setting F_PL
:
Setting F_PL
and dF_PL
:
Also we no longer see any sort of correlation between F_PL and DeltaG:
Without setting F_PL
:
While this is nice, I'm worried this is maybe too restrictive? Maybe we can just have a slightly more informed prior instead of this (link):
F_PL_guess = (Fmax - Fmin) / min(Pstated.max(), Lstated.max())
model['F_PL'] = pymc.Uniform('F_PL', lower=0.0, upper=2*Fmax/min(Pstated.max(),Lstated.max()), value=F_PL_guess)
Wow!
Can you show this to Nick? We might ask his thoughts on using an estimate of F_PL and its uncertainty from our spectra assay and/or a separate cuvette assay.