rickhelmus
commented
6 days ago Hi Theo,
Thanks for your interest and added details to your questions. I will try to answer everything in order, but please let me know if things need to be further clarified.
You are right the annotateSuspects() function is indeed used for ID level estimation, but currently limited to suspect screening workflows. However, it is relatively easy to use implement a similar approach. Also, the next major version of patRoon will also provide estimations for full NTA workflows. If you want to get a hint on how things currently work for suspects, please also see the default ID level configuration file.
Now to the numbered questions:
-
The score column is indeed defined as the weighted sum of all other scores. This is exactly as what MetFrag outputs, but additionally may include the 'formula score' if you ran the
addFormulaScoring()function. The scores are indeed normalized, unless MetFrag already did so (e.g.individualMoNAScore, see also thedefaultExclNormScores()function). The weights in patRoon are by default 1.0, to change them see e.g. the last example here. Since the defaults are unlike those in the Table you cited, and you might have usedaddFormulaScoring(), the total score you see may indeed by higher. -
There is currently no outlier detection. But this would be quite an interesting idea. Perhaps an approach could be to implement a function that re-calculates the score after e.g. using
delete()to manually remove some outliers. One complication, though, is that thescorecalculation from MetFrag is always from all candidates, while patRoon by default only loads the the top 100 per feature (configured by thetopMostparameter). Playing with weights may of course also be an option for some specific scoring terms. If you suspect the value ofdataSourcesis really off, it might als be good to double check the results from MetFrag directly, e.g. through the web interface, or have a look at CompTox online. If there is a difference, please let me know, as the values should be the same with patRoon. -
The way patRoon uses scores for ID level estimation you can find in the ID file that I linked before, which corresponds to the table you refer to. It doesn't use the MetFrag score directly (although you could change this if you want). The indivudualMoNA score of MetFrag represents the maximum spectral similarity of your experimental MS data and all the spectra for the candidate that are available in the MassBank of North America. So a score of 1.0 is a perfect fit. For zeros it simply means that the candidate is absent on MoNA (which unfortunately happens a lot!). For level 3c a similar approach is used but with silico data, i.e. the 'annotationSimilarity`: this is simply the spectrum similarity between your complete MS2 spectrum and your MS2 spectrum with only the annotated peaks.
I hope this clarifies things a bit, but don't hesitate to ask if you want to know more!
Thanks, Rick
Theodoralee27
Dear Rick and Team,
I would like to ask for help regarding how the scores for each feature are calculated from the generateCompounds() function. My main objective is to classify the features identified using Schymanski's level of confidence (Levels 2a - 5). Currently the most relevant function that links both the scores calculated in patRoon and Schymanski's confidence levels is the annotateSuspects function - but I realised this is only relevant for suspect screening, while the data im interested in classifying is the non-targeted data.
Here is a preview of what my data looks like after I have use the generateCompounds() function using metfrag and the Comptox database.
Here is a screen shot of their calculations:
I tried very briefly to calculate using the data I have, the resulting scores are pretty similar. I would like to confirm is that the correct calculation method? If not, can you kindly refer me to any document that discusses the calculation of the resultant scores?
My second question is a problem developed from the first - with the known max score of 9.0, I had two features in my dataset with a score of 13 and above. I was very confused which made me doubt the calculation methods for the scores, but when I went to investigate each parameter, I realised the value for 'DATA_SOURCES' for these two features were unusually high, around 40 - 42, while the rest of the other features only ranged about 1 - 8. Because of the way each parameter is normalised against the largest value, I was wondering if the system could recognise high outlier values like the 40 and 42, such that they are excluded from the analysis, otherwise all the other compounds scores will be very low for that parameter ("DATA_SOURCES") - again this is my speculation. I will upload one batch of data with that feature for your reference. Compounds_comptox_Npb6.csv
Is there an established conversion between metFrag scores and Schymanski's confidence levels? Unfortunately I havent been able to find much supporting evidence for this. The closest would be the use of the IndividualMoNA score > 0.9 to match a Schymanski level 2a (same study Lai et al 2021). Another study suggested a range of MoNA scores for levels 3a and 3b in Talavera Andújar, B. et al. (2022) ‘Studying the Parkinson’s disease metabolome and exposome in biological samples through different analytical and cheminformatics approaches: a pilot study’, Analytical and Bioanalytical Chemistry. Springer Berlin Heidelberg, 414(25), pp. 7399–7419. doi: 10.1007/s00216-022-04207-z.
But all my values for the IndividualMoNA for all my features were all 0 - would this mean that all my features are of level 3c or lower confidence?
Thank you so much for your help in advance! I would really appreciate if you can share any details regarding the score calculation!
Sincerely, Theo