Have a "refineFeatures" method

[jorainer]

Similar to the refineChromPeaks() method we could have a refineFeatures() method to clean/fix some correspondence issues. Again, would be nice to have "Param" classes to choose how to refine and specify the parameters. Possible refinement strategies that come to my mind would be to:

• merge features, e.g. if they are overlapping on RT and m/z - and based on some other criteria eventually.
• deisotope data - to remove features that are (clearly) isotopes of other features.

... ?

[JohanLassen]

Isn't this a flavour of what MsFeatures / CAMERA does? So by a few adjustments of the outputs of those packages, all we need is a "collapsing" method of the grouped features.

My current pipeline collapses MsFeatures groups by the mean values of the highest correlated feature pair within each group. By using rm.na=T when calculating the mean I also impute some missing values. Of course, this is agnostic to information about "main ions" etc. but the boost in ML performance using collapsing, is notably better than the effect of scaling/normalizing, imputing, and choice of ML method (within reason). By retaining the group information I can always backtrace the underlying features.

[jorainer]

Thanks for the feedback @JohanLassen . yes, I agree, there is overlap. This was also just an idea - but of course, we should rather focus on MsFeatures and implement the still missing functionality (a dedicated adduct/isotope grouping) there.