0
$\begingroup$

I have 4 protein microarray datasets that I am trying to compare. I have concocted a method of aligning them all and comparing them all on a continuous scale which seems to have worked well. These datasets have all been scaled individually using min/max normalisation and then plotted on the attached heatmap. X axis are all patients from the combined studies. Y axis is the position in the target protein. Z (colour) is signal intensity.

Happily, I did (visually) find a nice concordance between the datasets. However, as you may notice, there is a clear discrepancy between the 4 datasets, indicated by the visual colour partitioning on the plot. To address this, I have theorised that I might be able to increase the 'gain' on 3 of the respective 'low signal' datasets by taking the mean of each and iteratively linearly scaling them (adding a constant value to each measurement) until the means (total datapoints across each single dataset) match.

I'm not sure if this is a) a valid technique b) simply going to systematically increase background noise with no tangible gain to 'the important bits'.

In my mind would be another scaling method where the values added to each datapoint would be scaled more appropriately in themselves, reducing the overall addition of useless noise to every datapoint, and adding signal to the more meaningful datapoints. I know that sounds like a big ask, but I’d appreciate any help.

(I'm aware the right-most dataset is noisy af, but I can live with that.)

the heatmap

$\endgroup$

1 Answer 1

0
$\begingroup$

I tried using different scaling methods, such a robust scaling, normalising distributions etc. To no avail.

I did it. Turns out it was ridiculously simple.

I divided each value in respective datasets by the dataset total mean. This then gave each dataset a mean of 1, making the signal much more comparable. Best of all, I feel like this is a pretty valid technique, as this kind of transformation/scaling is common in serology inter assay comparison enter image description here

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.