rescale fluxes to logscale

[augeorge]

re-parameterizing the fluxes to log scale may help with the inference which can be slow if the scales are different by many orders of magnitude

[augeorge]

@djinnome @mcnaughtonadm

[mcnaughtonadm]

I believe we are already log scaling the fluxes (both the steady state calculated and observed) in the code.

It's commented out on main but in my working branch re-enabling the flux distributions we have:

log_vn_ss = pt.log(pt.clip(vn_ss[:, self.v_inds], 1e-8, 1e8))
log_vn_ss = pt.clip(log_vn_ss, -1.5, 1.5)

for the steady state and

 log_vn_obs = pm.Normal(
                "vn_obs",
                mu=log_vn_ss,
                sigma=0.1,
                observed=np.log(self.vn).clip(lower=-1.5, upper=1.5),
            )

for the observed fluxes. Is this what you meant by re-parameterizing to log-scale?

[djinnome]

Yes, that is what we meant.

@ShantMahserejian how are the eflux and enzyme data coming along?

[ShantMahserejian]

I just pushed the new model, notebook, and fluxes files. New flux file for the circadian experiments with the further refined model located at data/circadian_experiments/processed_data/enzyme_constrained_fluxes_nocofactorsinmodel.csv on branch #17-remove-ill-conditioned-metabolites-from-the-biomass-reaction. I'll start the pull-request now