#152

Ok, I'll take this from here.

Ok, made the changes here. I didn't do the notebooks, so please fix those. You just pass the time now, not a vector with the time in the second element.

Are we using any preallocation in ODEjac?
There is this error about that ODEjac cannot take Dual numbers:

I read that AD doesn't work with preallocated functions

Oops. Think I just fixed that.

diffCell still doesn't work

I found this:
link
It looks like there is nothing to work around it

Is this way more accurate than calculating all the points?
The behavior is less monotonic.

Post plot?

dox:

gemcitabine:

lapatinib:

paclitaxel:

Hmm... the output should be the same. I did change the time since you were indicating the 100th timepoint before, and now it's 100.0 hrs. Maybe you could figure out what time the 100th point is, and then set it to that?

Also adding more concentrations may help find the issue. Because there's oscillatory behavior the result doesn't have to be monotonic, but it should be locally smooth.

Of course. Thank you.

The animation looks cool! This is the plot for almost all times.

Could I keep it in the notebooks?

@aarmey I couldn't figure out what the problem is. Could you please let me know what I did wrong or what I should do to correct this?

This is related to my post on #general. Should be fixed now, but juptyerlab won't be available until I hear back from seasnet.

Thanks.

Sometimes the cell number becomes negative. I put an assertion for that, I will try to see if this used to happen before the change to the jacobian.

I guess there is something wrong with the Calculus.finite_difference() because as it is clear here, with exactly the same set of parameters, the number of cells is decreasing.

The y-axis is the number of cells, and the x-axis is a range of gamma_1.

@aarmey do you see any other point that might be going wrong here?

Something that looks weird is that with a higher concentration (this is in gemcitabine) the G2 rate progression -- which also happens to be division rate -- increases! I searched a little to see if I could find some evidence of this, but couldn't.

Actually, there is something wrong with the predict2(), it sometimes returns negative cell numbers.

Since the solution in predict() and predict2() is not the same, and the second one sometimes returns negative cell numbers, if you let me I will use the predict() and just take 1 time point. Is that alright?

Yep. In that case get rid of predict2

Some updates:

1. paclitaxel and doxorubicin seem to work alright in terms of their gradient. Gemcitabine goes positive for two last concentrations. There should be something related to the fact that beta increases significantly over concentrations.

There are two errors now.

1. For all of the sensitivity analyses, the only first cell # is some negative small number (like -0.002). I changed the range we have for the parameters in sensitivity analysis, but it just happens with any range.
2. In lapatinib, the number of death phase steps in G1 (nD1) is estimated to be zero. This messes with something that it errors for invalid array dimensions. It occurs when it reaches the line here:

This is the error regarding invalid array dimensions.

@aarmey I can't figure out what array dimension is wrong. Could you pleeease take a look at this error? When nD1 is estimated to be 0 for lapatinib, this error happens. I mean when the matrix dimension for death in G1 becomes zero. The fitie_difference breaks in this. I tried a few tricks to just remove that parameter, but it breaks other things. I would really appreciate if you could take a look at this. Any recommendations helps.

If we're not ultimately using predict2, then what do we need to keep from this branch?

I was trying to address the issues in the gradient here, but you are right, I could do it in another branch.

This error I mentioned still exists even with what it is in the branch master. It is for the control trial in lapatinib.

Yes, if there's nothing you need here, it'd be helpful for debugging to make a failing case off of master.