nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Bioassay analysis - Log4p model with base 2 instead of e

martindemel

Hi,

my customer (Sandoz) asked if there is an option to use custom functions in Fit Curve for bioassay analysis and parallism test / equivalence test.

“Regarding the bioassay you showed in the beginning. Is it possible to create a user specified 4PL model for curve fitting of bioassay data? In JMP 4PL model uses power with base "e", while we use a model with base "b", where b = 2. I guess we solve this with a JMP script? Is it possible to also perform parallelism and equivalence testing like in a pre-defined model?”

A bit more elaborated:

The reason for using a different base is that we rely on validated software to calculate the relative potencies of bioassays. The model implemented in that software uses a base of 2, or at least that is how it was specified at the start of the project.

My intention would be to use JMP primarily as a screening and exploratory tool. In the validated software, every action is recorded in the audit trail, which makes exploratory analyses rather time consuming. For example, we are sometimes interested in removing a specific dilution point to assess whether suitability criteria such as equivalence, parallelism, R^2, and others improve, or to evaluate the impact of that change on the relative potency estimate and its confidence intervals.

Performing these investigations in JMP would allow us to quickly assess how small changes in the data influence the overall result. It would also significantly speed up the evaluation process. I see JMP as a useful sandbox environment where changes can be tested immediately and the effects reviewed in real time.

As mentioned, we currently perform this work in the validated software, but each modification requires documentation and justification. I have already tested the approach in JMP, and it works well. The only difference is the model base, which is why I reached out.

I could not find a way to change the base or use a custom function in Fit Curve. I know this would be possible with nonlinear platform but then there is no parallism/equivalence test.

I think - taking into account the context of the validated system - it might be an option to think about adding a custom option as well to the upcoming bioassay branch in Fit Curve. Though, not sure how that would affect the parallism and equivalence tests, aka are ther assumptions which cannot be assured if there is an option to change base or use custom functions.

@Valerie_Nedbal

Accepted answers

All comments

Jed_Campbell

If you save the parameterized prediction formula from the Fit Curve platform, you can use that same formula in the Nonlinear platform, and you can modify it. If I'm understanding correctly, this would give them what they want. Plus, the Nonlinear platform's parameterized formulas have sliders and optimizers that make exploration fast/interesting.

martindemel

Hi Jed,

correct me if I'm wrong, this would allow them to change the formula and see how this will fit, but they would still not be able to perform a parallelism test or equivalence tests based on this new formula, and that's what they are wanting to do, getting these reports just for a different/custom function rather than the build-in ones.

The approach in the nonlinear platform will allow them manually to extract inflection points through the profiler for each curve manually and then compare those values. But they would not get the reports automatically as it is in Fit Curve.

martindemel

Here is the wish list entry for completeness:
https://community.jmp.com/t5/JMP-Wish-List/Custom-4PL-Models-for-Exploratory-Bioassay-Analysis/idc-p/961041#M8609

clay_barker

Thanks for sharing this one, Martin. Adding the ability for a user to specify custom models to Fit Curve has come up several times, but is kind of beyond the scope of the platform.

Adding new built-in models to the platform can be relatively straightforward though. Just making sure, is this the form of the model that the customer would like for us to add?

martindemel

@clay_barker I just came back from vacation and checked back with the customer, here is his response:

This is the 4-PL model we use, very similar to what you posted:

where small b = 2. The idea was to allow users to choose the value of b freely.

Sorry for the different letter notations, but I think it is still clear. Though they currently use a base b=2 , they would like to be able to chose the base freely.

martindemel

Might be an idea to implement this general model with a default base and an option to change the base (similar to how we change alphas and other parameters in other platforms)?

Byron_JMP

I'd bet the customer is using something like graph pad prism. That program uses log10, and its iterative solution is just a little different.

its super easy to make a new formula column and log 10 transform the X column. The fit parallelism > relative potency assumes that the x is log10. you could use ln or really, any other transform, but each one will come up with a different answer because the scale is different. in bioassays log10 is the very most common X col transformation. I've heard of someone trying to use log2, but of course none of their results matched anything from the literature.