Feature request: Logistic fit function for quantification

support
Feature request: Logistic fit function for quantification MT  2025-05-23 02:45
 
Dear Skyline-Team,

I hope this is the right place for my inquiry.

I do absolute peptide quantifications on TQs and Orbitraps using isotope labeled heavy internal standards and calibration curves.
The workflow contains an immunoenrichment step, which is why antibody kinetics have a great influence on the calibration curve shape (usually sigmoidal even on log scale).

Currently, I use Skyline only for the peak integration; quantification and evaluation is subsequently done in Excel and Origin. I would love to do the quantification with Skyline but, as far as I know, Skyline does not support a logistic fit model. Would it be possible to implement this kind of fit in the quantification tab?
Most vendor software (including SciexOS, TraceFinder, and XCalibur) have only linear fit models available, thus this is a real need in the IA-LC-MS/MS community.

Thank you in advance!

Best regards,
Tobias
 
 
Nick Shulman responded:  2025-05-23 09:34
On the "Quantification" tab at "Settings > Peptide Settings", you can choose "Bilinear" in the "Regression Fit" dropdown.
When you choose "Bilinear", Skyline fits a piecewise linear equation to the data. That equation has a horizontal part, and a part with a positive slope.
This piecewise linear fit looks similar to the lower half of a sigmoid function.
I imagine you would need to have a lot of very precisely measured calibration points in order to be able to distinguish the lower half of a sigmoid from Skyline's bilinear fit.

Is the bilinear regression good enough? Or is it important to model the upper half of the sigmoid where it goes flat again above the upper limit of quantification?
-- Nick
 
MT responded:  2025-05-26 01:00
Dear Nick,

Thank you for your quick reply.

Similar to ELISA, the full sigmoidal curve can be used for quantification, as long as the method has been properly validated, a suitable fit model has been used, and the calibrators fall within their accuracy acceptance criteria (usually 80-120%).
I agree, that often only the lower part is of interest, but depending on the analyte and the assay, it is possible that the full curve is required.

I will check if the bilinear regression can be used for some of our analytes, i.e. those where only the lower part of the sigmoidal curve matters.

Are there other options than the bilinear regression?

Best,
Tobias