Closed jbloom closed 5 months ago
For fixslope, I initially tested [0.8-6] as the starting range, as for the DRIVE cohort data, where this parameter was not constrained, this range appeared to capture the majority of well-fit curves (~65%).
I also tested [0.8,10] with the most recent data, and I do have a large number of curves fitting with the max slope within the range permitted (either 6 or 10, respectively), but I think that may be partially a result how the error in the most recent plates is impacting the shape of these curves, i.e. we have many curves with no points in the slope or only a single point in the slope. The resulting midpoint titers are quite similar, as would be expected. I would expect that the larger this max slope is the more variability we might observe between slopes fit for samples with sharp curves like this, so a smaller range is likely preferable. Though, I will repeat this test once we have additional plates to confirm that these parameters are reasonable.
Similar to @anloes, I have also been constraining fixslope
to a fairly loose range of [0.8, 10]
. I agree that the difference between midpoints called with [0.8, 10]
versus [0.8, 5]
is likely to be minimal, but for now, I will probably continue to use this wider range until I have more data. I'm currently only testing with 9 individuals.
In the future, since there's no real biological reason (?) to point to for allowing such steep slopes, I agree that reducing the upper bound is probably better.
In #35, it was implemented to enable
fixslope
to constrain the slope to a range (see #32).However, more analysis of real data is needed to see if the suggested range in the example (
[0.8, 5]
) is actually reasonable for serum.Check this on real data and adjust accordingly.