Refit IL-2 models
Closed this issue · 9 comments
Idk if you've already started to re-fit but it might be worth it to wait until the Julia PR is merged with master. It will probably be faster solving the ODEs and at the very least we can say we were consistent with our solver implementation in the manuscript.
Think the Julia PR might not finish until after we've submitted. A few parts have turned out to be tricky.
And, actually, I think it'll be about the same speed or very slightly slower—it's still going to run through Sundials/CVODES. I'm hoping we can make the change at some point because it reduces the solver code by ~1000 lines, and I think will be easier to change in the future. We can potentially do future fitting work in Turing.jl, which has a much simpler syntax than Theano.
Also, we'll have to only make a fork of this repository public—we now have non-public data on the Visterra molecules which will have to stay private.
Gotcha. I guess I wrongly assumed that fewer lines of code translates to more efficient back-end implementations of the Sundials/CVODE package... Let me know if you want my help with this. It should be as easy as executing the fit command again (with proper arguments) but nevertheless I can keep track of this so you can focus on other (less trivial) tasks.
Holding off on this until we have finalized how we handle YT-1 receptor measurements.
@adamcweiner this is ready for you—we can re-fit the IL-2/15 models. Note that we're going to submit without the YT-1 receptor quantitation. Please take out the gc measurement before doing so.
Running both fits (w/ & w/o trafficking) now. If I'm using too much comp power by running both at the same time feel free to cancel one of them and I can re-run later.