How Replicates are Treated in the Functional Data Explorer

MedianRooster42

Hello,

I have functional data I have entered into the functional data explorer. I have three biological replicates of absorbance data over time, which shows the growth of an organism in triplicate over time using an absorbance measure. This is dense data, and I have measurements in each replicate every 30 minutes for 24 hours. This means for a given organism, I have three absorbance measurements being taken at the same time point, every 30 minutes. I have many organisms, and I want to run an FPCA to cluster them based on the absorbance functions.

I was wondering how JMP and the functional data explorer (p-splines through to PFCA) treats replicates entered as the same ID? When the data is modeled, I can see the models almost take an average of my three replicates (see attached).

What exactly is happening? I am giving JMP three x-values and three y-values at each 30 minute interval, so surely there would be some sort of standard treatment when doing so, as this almost seems like it should not be a possible input for JMP?

Thanks in advance for your help.

Screenshot 2026-07-01 at 7.55.15 PM.png

Screenshot 2026-07-01 at 7.55.20 PM.png

Victor_G

Hi @MedianRooster42,

Welcome in the Community ! 

If your replicates have the same ID, the Functional Data Explorer will consider these runs as the same experiment, and model the results through an average function (or try to interpolate through the points if you had a different timestamp/delay measurement between experiments). Or said differently in the JMP Help: "For Stacked Data, if no ID variable is assigned all observations are assumed to come from one function." (and same situation if one ID variable is assigned to several observations, it is assumed they come from one function).

Replicate runs are independant runs, so I would give them a different ID. Maybe you could think of a coding like Experiment A-1, Experiment A-2, Experiment A-3, ... or something similar.
It makes sense to provide a different ID from an experimental point of view (independant runs) as well as modeling point of view: you keep all raw data without any aggregation, and the functional model will be fitted through all experiments, no matter if they are replicates or not. This way, you can assess the variability in your replicates by analyzing the Function Summaries for your replicate runs, and particularly the FPCs scores. Analyzing the replicates this way may also make more sense if you intend to use Functional DOE Analysis : Example of Functional DOE Analysis

Hope this answer will help you,