search

MassBank / RMassBank

Playground for experiments on the official http://bioconductor.org/packages/devel/bioc/html/RMassBank.html
Other
12 stars 15 forks source link

Debug repeated peaks with no (yet) apparent cause #53

Open sneumann opened 10 years ago

sneumann commented 10 years ago

Issue by schymane from Friday Jan 10, 2014 at 10:37 GMT Originally opened as https://github.com/sneumann/RMassBank/issues/45

Examples 278 (pH_120120) - the spectra are a mess! Reacquisition needed?? This is currently excluded from all massbanks... 124 (pH_120220) - the spectra are a mess too, but pH_120108 worked fine.

schymane commented 8 years ago

I found a case where I get repeated peaks and can put it down to a very poor recalibration because I don't have enough points. The problem disappears if I take the recalibration from another mode or if I recalibrate.identity. @ermueller and @meowcat have some files. I don't know if this is the cause in all cases? Do we need a warning on the recalibration if it's crazy like this (because if people keep going the results will be awful...)

schymane commented 8 years ago

From @meowcat I examined the error, it has a really curious cause: out of apparently pure coincidence, the recalibration curve maps three masses directly onto each other!

One might catch this for example with if(any(abs(drecal) > 100) warning("crap recalibration")) in recalibrateSingleSpec. (which would warn on recalibrating anything by >100 ppm. In this specific case the mass was recalibrated by 50 000 ppm :) )

meowcat commented 4 years ago

@schymane @tsufz Any reports of this still happening? Test case? - Other than that, it would be useful to highlight when the recalibration curve has problems, e.g. NA in the return values, huge steps, not enough data points, and to tell the user to use identity or use more data.

schymane commented 4 years ago

No test case available easily at this stage but I agree that this:

Other than that, it would be useful to highlight when the recalibration curve has problems, e.g. NA in the return values, huge steps, not enough data points, and to tell the user to use identity or use more data.

would be very useful... Hiba encountered terrible recalibration issues. CCing @anjuraj15 as maybe they can provide some examples. Maybe we should open a new issue for that @meowcat ?