Clarification needed on "GC-EI spectral deconvolution"

[Sanchezillana]

Basic information

• My operating system and version: Windows 11
• My mzmine version: mzmine 4.0.8

What happened

I am currently working with the GC-MS workflow in mzmine and have a doubt regarding the type of deconvolution implemented in the module "GC-EI spectral deconvolution".

I have searched through the documentation but couldn't find detailed information on the algorithm used. Could you please clarify whether the deconvolution in this module is based on hierarchical clustering, MCR (Multivariate Curve Resolution), or another method?

Your assistance would be greatly appreciated.

Thank you!

[Sanchezillana]

In fact, we have the same question for GC aligner. Is it ADAP-based or HCA-based?

[ansgarkorf]

Hi,
the documentation is currently missing, sorry for that. The module was recently introduced and does not use hierarchical clustering nor multivariate curve resolution.
Can you elaborate on your doubts regarding the type of deconvolution and let us know what you would like to see?

[Sanchezillana]

Hi, the documentation is currently missing, sorry for that. The module was recently introduced and does not use hierarchical clustering nor multivariate curve resolution. Can you elaborate on your doubts regarding the type of deconvolution and let us know what you would like to see?

Thank you for the information. I'd like to know more about the type of deconvolution algorithm used in the new modules "GC-EI spectral deconvolution" and GC "aligner", including specific details about the algorithm and any references to literature or related work. Additionally, it would be beneficial to receive some feedback on the configuration parameters available, such as the default or fixed parameters. This information would greatly help in understanding and utilizing the module more effectively not as a black box. Thank you!

[ansgarkorf]

#1916

[Sanchezillana]

#1916

Congratulations on the module update! It looks fantastic. I have still a couple of questions about the GC-MS workflow:

• Could you provide literature references and details about the deconvolution algorithms used in this module?

• How is the intensity (or area) computed in the final feature table of the GC-MS processing workflow? Specifically, is the signal derived from the most intense peak in the deconvolution, or from the somehow the entire deconvoluted profile?
  In my experience, implementations like Wehrens' metaMS and Domingo-Almenara's eRah work quite well. It would be great to have the option to select different algorithms for better flexibility.

Thank you!

[ansgarkorf]

Hi,
thanks for the comments. The PR is work in progress and your feedback is very helpful.

The module will be designed to use different algorithms that the user can select. Literature sources (if available) will be added to the documentation.

Thanks for sharing your opinion on Wehrens' metaMS and Domingo-Almenara's eRah. I will have a look. Currently intensity and area are derived from the most abundant grouped feature, indicated by the peak area in the preview. The idea is now to add an extra module that can be run optionally to also create a deconvoluted feature. What do you think about this approach?

[Sanchezillana]

Hi, thanks for the comments. The PR is work in progress and your feedback is very helpful.

The module will be designed to use different algorithms that the user can select. Literature sources (if available) will be added to the documentation.

Thanks for sharing your opinion on Wehrens' metaMS and Domingo-Almenara's eRah. I will have a look. Currently intensity and area are derived from the most abundant grouped feature, indicated by the peak area in the preview. The idea is now to add an extra module that can be run optionally to also create a deconvoluted feature. What do you think about this approach?

This sounds great! I believe this is the best approach. If I understand correctly, this deconvoluted feature will have an area or intensity proportional to the concentration of the feature somehow. Do you have any idea how are you going to implement the computation of this signal?

Regarding the current approach based on the most intense mz, I think it can be problematic when this mz is a low-mass peak prone to saturation or more irreproducible. The regression approaches seems like a better method. However, there might be more advanced approaches available now. We would be happy to collaborate on this. We're using data beyond metabolomics, such as fuel analysis, and I believe mzmine could be very powerful in these GC-MS applications as well.

[ansgarkorf]

Hi again,
first part of the GC updates was merged. You can test it in the development build:
https://github.com/mzmine/mzmine/releases/tag/Development-release

Documentation can be found here:
https://mzmine.github.io/mzmine_documentation/module_docs/featdet_spectraldeconvolutiongc/spectraldeconvolutiongc.html

Looking for more helpful input, especially on determining a good model feature.

[Sanchezillana]

Hi again, first part of the GC updates was merged. You can test it in the development build: https://github.com/mzmine/mzmine/releases/tag/Development-release

Documentation can be found here: https://mzmine.github.io/mzmine_documentation/module_docs/featdet_spectraldeconvolutiongc/spectraldeconvolutiongc.html

Looking for more helpful input, especially on determining a good model feature.

Really nice! I will look into it. What about the last reference of GC-ADAP? https://pubs.acs.org/doi/10.1021/acs.analchem.9b01424
Is it the same implementation?

[ansgarkorf]

Not the same, but similar logic. ADAP will also come back at some point.