Some questions about single peptidoform prediction

[ypriverol]

@RalfG :

We have been trying to use MS2PIP in combination with USI. The use case is that for a given USI get the predicted b,y ions, retention time, intensities, etc, would be good to have also correlations.

I planned to use the ms2pip.predict_single, but this function only predicts b,y ions for a given peptidoform. However, would be possible to extend that function for:

• peptidoform,
• retention time
• experimental peak information.

In the current implementation, I have to take the given peaks and write them down into a file, to use correlate I guess?

[RalfG]

Hi Yasset,

Good point! We hadn't considered this usage mode yet, as it is not possible to do this through the CLI. (Pretty difficult to pass an observed spectrum through a CLI argument).

I added a new correlate-single usage mode in #232 for the Python API only. This should be part of the next release. In the meantime, you can use the code in this Colab notebook. It takes a USI, MS²PIP model, and mass tolerance in Da, then gets the spectrum through the Pyteomics ProXI client, annotates it with MS²PIP, gets the MS²PIP predictions, calculate the Pearson correlation, and as a bonus, plots both spectra with spectrum_utils.

I'm not sure how to add retention time to this comparison, as this would require knowledge of the complete MS run with at least a some other PSMs for RT calibration.

Let me know if this fits your needs, or if you have questions.

Best,
Ralf

[ypriverol]

Thanks, @RalfG that will do the work for me. Im making an API endpoint that does exactly that. Given a USI and including peptidoform give you back the predicted intensities and annotations.

@RalfG Im quickly checking the Notebook, Im just wondering, why you need these two lines of code:

 observed_spectrum.tic_norm()
 observed_spectrum.log2_transform()

[RalfG]

Perfect!

The tic_norm method normalizes the intensities to the total ion current (TIC), similar to a base-peak transformation. The log2_transform method takes the base-2 logarithm of the intensities with an addition of 0.001 to avoid divisions by zero (sometimes called epsilon or fuzz factor).

The former is done simply to get a consistent input, the latter to make the distribution of intensities more linear and therefore easier to accurately correlate. Differently put, it de-emphasizes the higher peaks and emphasizes the lower peaks. TIC-norm and log2-transform were found to give the best similarity measure with Pearson correlation in the following work: https://doi.org/10.1002/pmic.201000605. Nevertheless, I expect very similar results with base-peak normalization and square root transformation.