When executing experiments, there are cases where variation occurs across a multiple sets of data in your experiment. These sets of data are called experiment 'blocks' in statistical parlance, but you likely think of them differently (and more intuitively) as:
- Sets of samples run across multiple multi-well microtiter plates (the plates are the blocks).
- Sets of samples measured across multiple days (the days are the blocks).
- Sets of samples measured across multiple sites (the sites are the blocks).
- Sets of samples measured across multiple lots of media (the lots of media are the blocks).
If you can describe any part of your data as "sets of samples measured across multiple ___" then you might have an experimental blocks. Blocks can lead to undesired variation in your sample measurements. For example, an offset in the baseline assay value for samples on one microtiter plate versus another.
This sort of variation is noticed when reference standards are included in the blocks. If those standards are showing systematic assay value variations from block to block, that means you have systematic errors.
Experimental error is obviously highly undesirable, but the good news is that you can easily correct this error during analysis. As long as you have reference standards shared across multiple blocks, then you can remove this error during data analysis. You simply use the measurements of the shared reference standards to mathematically calibrate (correct) the measurements of your unknown samples.
Even better news — the Riffyn Measurement Normalization Tool does this for you automatically! Click, click, click and voila! Clean data. The Measurement Normalization Tool is in the JMP add-in found here.
How it works
The Measurement Normalization Tool uses the "Fit Model" platform in JMP to adjust measurement data based on the effects of your blocks (a.k.a. "blocking factors"). To use this tool, open your data table (a sample data set is available at the end of this document) and select the option from the JMP menu:
This will open a window for you to select your measurement value, the blocking factor(s), and the sample identifiers. In the case of our sample data set, we want to investigate plate effects (plate ID column).
After we have filled out this section we can explore the model that will be built and add/subtract model effects based on the results we see by clicking on "preview model". Model modifications are generally something you want to do only if you have a deeper understanding of linear regression principles, but it can be very useful and important. So don't be afraid to poke and prod to learn how it works.
After clicking the "Normalize data" button, the corrected data saved to a new column in your original JMP data table. It's called "<Property Name> - Normalized". <Property Name> is the name of the measurement data you selected at the beginning when you opened the normalization tool.
In our example experiment, we expect that the reference samples behave the same across each plate. But based on our raw measurement results, it looks like the plate is causing the reference results to vary (as shown by the red, blue, green stars in the top row of plots). Notice how the average values of the unknowns (purple - upper plot) also change from plate to plate in the top row. After correcting for the plate effects with the Riffyn tool, we can see that the data have been corrected such that we get the same measured value for each reference regardless of the plate (lower row of plots) .
In addition, to correcting the reference standards, the tool is also correcting the unknown sample measurement values by the same formula as was used on the references. Thus your data become "normalized" and the overall accuracy and precision of your results are improved.