Arseha / peakonly

MIT License
110 stars 37 forks source link

Congrats ! Suggestion interoperability / peaks grouping #1

Open lfnothias opened 4 years ago

lfnothias commented 4 years ago

Hi Peakonly community ! Thanks a lot for this amazing tool ! Data processing is difficult and very much user/parameter/tool dependent. With peakonly we can hope for simplified/fast and comprehensive LC-MS feature detection. Currently, results are not easily interoperable with other metabolomics tools. Would it be possible to have an mzTab-M export for the results ? https://pubs.acs.org/doi/abs/10.1021/acs.analchem.8b04310 Also it would make sense to find a way to integrate tools such as CAMERA or RAMClust to group peaks belonging to one compound, in order to reduce the data complexity and make the results usable. https://pubs.acs.org/doi/abs/10.1021/ac202450g https://www.mdpi.com/2218-1989/6/4/37/htm This could be done with ADAP potentially or even OpenMS FeatureFinderMetabolomics tools. https://pubs.acs.org/doi/abs/10.1021/acs.analchem.6b02222

Thanks again,

Louis

lfnothias commented 4 years ago

Actually, maybe the best/easiest solution would be using XNet, a parameter-less clustering algorithm made by one of the leading group in the field: https://github.com/optimusmoose/XNet https://pubs.acs.org/doi/pdf/10.1021/acs.jproteome.9b00068

lfnothias commented 4 years ago

Instead of mzTab-M (or along with), it could be a good idea to use the OpenMS .featureXML for the output. Then it makes possible to use any of their tool to build complex and versatile pipelines.

Arseha commented 4 years ago

Thank you very much for your suggestion. We will definitely add the possibility to export results in different formats.

Also it would make sense to find a way to integrate tools such as CAMERA or RAMClust to group peaks belonging to one compound, in order to reduce the data complexity and make the results usable.

In fact, we already have an implementation of an algorithm partially similar to RAMClust. Unfortunately, we cannot guarantee that it would be integrated very soon. We still have a lot of work to do before it (currently working on a very interesting approach for peak matching between samples, which probably could even increase the final "precision" and "recall" of the raw data processing), but we'll try to do it as soon as possible.

Actually, maybe the best/easiest solution would be using XNet, a parameter-less clustering algorithm made by one of the leading group in the field: https://github.com/optimusmoose/XNet https://pubs.acs.org/doi/pdf/10.1021/acs.jproteome.9b00068

It does sound like an interesting solution. We'll experiment with this approach and probably we'll also find a way to integrate it. Thanks again for the profound articles.

lfnothias commented 4 years ago

That sounds great ! I am happy to test what ever you developed on some data I am familiar with so I can provide additional feedback.

ankitshah009 commented 4 years ago

+1 to the above development and would be able to provide feedback on the current developed methods.