meyer-lab / gc-valent Goto Github PK

Bottom four rows of approximate jacobian are zero, indicating receptors are not responsive to endosomal ligand.

Currently the number of species and parameters is set independently wherever it's used. This means that when these change, there are ~20 places that have to be changed and matched up. If there's a constant in, say, model.py, this can just be imported wherever needed.

Shares beta and gamma subunits with IL-2.

We should check for:

• Detailed balance holding (i.e. model comes to equilibrium no matter parameters)
• Mass balance of individual receptors

Along with appropriate testing.

As there's increasing amounts of IL2 outside the cell, the endosomal concentration of free cytokine should go up as well. We should test for this, and that the other cytokines stay at 0.

We used dot to look at a single cytokine case—is there something we can use to plot all of the cytokine species?

Implementation in matlab available here: http://www.models.life.ku.dk/~rasmus/presentations/Npls_sugar/npls.htm

k5rev / kfwd = 1.5 nM according to (https://doi.org/10.1016/j.jmb.2004.04.038)

Solving is failing for certain parameters, with the corrector convergence test failed repeatedly or with |h| = hmin. This is being fixed on the runCkine_test branch.

The cytokine-gc complexes seem to be the issue when solving, as their reverse rates are high and so create a stiff ODE model. We can just assume that they're negligible, and assume when a cytokine-RA-gc complex comes apart that it completely breaks apart into constituent pieces. This will require changes in the code across many different parts.

Each time parafac is run different alignments of component 1 vs component 2 arise.

These will likely be SPR measurements, but could just be equilibrium binding. A place to start would be references in papers that refer to particular receptors as high or low affinity.

It looks like, in some of the unit tests, test-specific variables are being stored class-wide (i.e. with self.X). The self.X syntax should only be used when a variable will be used by more than one function. In the case of unit tests this should really only be used by variables that are setup in the setUp constructor. Please change the wrapper tests to use local variables unless you need these from setUp.

Currently NUTS initialization runs, but then fails during sampling with:

ValueError: Bad initial energy: nan. The model might be misspecified.

Probably some information on this here:

https://discourse.pymc.io/t/how-to-debug-bad-initial-energy-nan/492

Rather than making the factorization matrix for all possible levels of receptor expression, we should see if we can find particular combinations of receptors that are expressed in different cell populations. Not all combinations may be biologically meaningful.

This will come after finishing implementation.

I.e. set values of IL2, k1fwd, k4fwd, k5rev, k6rev, k10rev, k11rev and an initial y0 in the setup function. Save the values as member variables, so the tests can use them later.

Seems to be much more biophysical data available for IL4...

May offer considerable speedup.

Based on the Ring et al manuscript which suggests this binding is weaker than in the IL2 case.

https://github.com/meyer-lab/type-I-ckine-model/blob/14fd246ce2e2e5d025d7e41cab721f226589f57a/ckine/model.py#L40

Solving for the numerical Jacobian is available here:

https://mail.scipy.org/pipermail/scipy-user/2008-November/018725.html

Along with appropriate testing.