search

eugenemel / maven

Maven GUI: Metabolomics Analysis and Visualization Engine
https://github.com/eugenemel/maven/releases
GNU General Public License v3.0
19 stars 9 forks source link

Fragmentation scoring producing bad values #31

Closed PMSeitzer closed 5 years ago

PMSeitzer commented 5 years ago

Identified by Bryson in Maven 8.0.18. See screenshot. frag_scores_off

PMSeitzer commented 5 years ago

The CompoundSearch widget appears to be disconnected from the fragmentation scoring associated with the full database search.

Additionally, choosing the numRefMatches option in the CompoundSearch causes the program to crash.

PMSeitzer commented 5 years ago

It appears that this is a display problem with the Fragmentation Spectra plot.

The plot seems to be displaying the spectral peaks for the [M-H]- form of PI(18:0,24:0) instead of the [M+NaOCH-H]- form (as expected).

Somehow the proper adduct information is not being delivered to the fragmentation plot. Probably, the fragmentation plot is falling back to hard-coded adduct constants instead of the user-specified adducts list.

I will incorporate this work in Issue #26 in reforming the fragmentation score plot.

brysonbennett commented 5 years ago

I'm not quite clear. There is no adduct of type [M+NaO2CH-H]- in our database/msp files. Matching against the [M-H]- is functioning as intended for Maven (As maven calculates the masses of the adducts to check against based on ADDUCTS.csv). It should score that to a good score, if behaving as intended. There are also spectra for e.g. [M+FA-H] in the compound list, they just are not scoring as well. Fundamentally, I think this is just a problem in correctly calculating scores between the experimental spectra and the options in the databases that are loaded.

I'm not quite clear what mzkit is doing to score these, as it is obviously handling the M+O2CH-H adduct type (which I did not think it did).

PMSeitzer commented 5 years ago

The problem is more complicated, with some spectral matches scoring incorrectly.

PMSeitzer commented 5 years ago

To reproduce: spectral library, use 20190517-Calico-Lipids-neg.msp Dataset: DOMouseReduced folder contains the mzroll object and files

PMSeitzer commented 5 years ago

Find Peaks button is completely broken - crashes right away

PMSeitzer commented 5 years ago

Consider restructuring PeakDetector search into a single dialog?