Co-elution count

Co-elution count mvillal1  2017-08-14
Hello Skyline team,

A quick question, in the MProphet Features export option, how is exactly the "co-elution count" feature being calculated? Does it take into account the # of co-eluting peaks having the same or similar apexes (falling within certain delta RT) or is it based on the "co-elution" score (cross-correlation)? In some cases, after doing an automatic selection of the top 5 transitions per peptides, the number of co-eluting transitions goes up (i.e., from 2 to 4) even when the non-refined list included those transitions. Also, the feature just seems to report a max. of 4 co-eluting transitions.

Many thanks!
Nick Shulman responded:  2017-08-17
Yes, the coelution count is always going to be a value between 0 and 4.
If the total number of transitions for the peptide is greater than 4, then the formula is:
4 * (# of transitions that had a peak) / (total number of transitions)

if the total number of transitions for the the peptide is less than or equal to 4, then the formula is:
(# of transitions that had a peak).

Here is where the source code for that calculation is:
It's in the class "LegacyUnforcedCountScoreCalc", but if you look at line 102 ("GetPeakCountScore") is where the multiplying by 4 and dividing by the total number happens.

I am not sure what the exact definition of a transition "has a peak", but I believe it is the same as a transition that would have a missing value if you were to uncheck the menu item "Settings > Integrate All".
mvillal1 responded:  2017-08-19
Thanks for the reply Nick.

Attached is a ppt slide with two peak groups (without the "Integrate All" setting selected but all integrated or showing green icons on the main document) identified in the same XIC with different "co-elution count" values. For the case at 37.5 mins, which transitions are co-eluting according to Skyline? Since some of the values for "co-elution count" being reported in the MProphet Feature output are not exact integers (i.e., 3.3 in this case)... Does this mean that they come from using the formula? If this the case, is the selection of co-eluting transitions based on a specific RT threshold at the peak apex?
Nick Shulman responded:  2017-08-21
I believe the best way to figure out which transitions Skyline think have a peak there is to look at the colored dot next to the Transition in the Targets tree.
I think you also have to turn off "Settings > Integrate All", because I think when "Integrate All" is turned on, all transitions get a green dot, I think.

I do not know whether we have written down anywhere what the definition of a coeluting peak is.

The code for that is in a function called "FindCoelutingPeaks" in this file:

That code includes criteria such as the transition needs to have a peak at least 1% of the size of peak of the largest transition at that time, or that apex of the smaller peak has to be within 25% of a peak width from the apex of the larger peak.
Brendan MacLean responded:  2017-08-21
Well, it's complicated. Skyline considers a maximum of 10 peaks in any replicate for any peptide. These all get stored with a flag for each transition indicating whether its chromatogram in the peak's time range appeared to co-elute with the dominant peak around which the peak group was assembled. The capping at 4 helped to normalize the score and not overweight targets with a lot of transitions, but it is also clearly very arbitrary. If your document has targets all with the same number of transitions (as many do), this normalization will not make any difference. The score was part of the original Skyline default score before we implemented mProphet. After implementing mProphet we found it usually ended up strongly weighted in the trained mProphet models, and so we kept the score as it was.

So, which transitions are flagged and which are not?

As Nick indicates, if you turn off Settings > Integrate All, then Skyline will show the unflagged transitions with red dots in the Targets view... but only for the peaks it selects by default. All manually chosen peaks, whether you chose them by dragging a range beneath the x-axis or simply by clicking on the retention time annotation above the peak behave as if "integrate all" were turned on. (Note: when integrate all is on, the only transitions that appear with red dots are ones that contribute no peak area to the total area.)

I recently found the loss of co-elution information in manually changing peak picks a little frustrating myself, as I was trying to understand why Skyline made a surprising peak choice, and this made it harder for me to understand that the very nice peak Skyline failed to pick had ended up with fewer "co-eluting" transitions than another it picked instead.

I have been thinking about ways to make it easier to understand, within the Skyline UI, the peak scoring and picking. To that end, I am planning on implementing some way that you can get a pop-up tip detailing the various feature scores, the weights used and how they add up to the total for each peak, when you hover the mouse cursor over the peak retention time annotation (probably some option in Tools > Options to turn this on). Based on your interest and my recent frustration, I will also try to get implemented a right-click menu when clicking on a peak retention time annotation that has options like (Integrate All and Integrate Co-eluting) to allow a peak-choice that preserves the flag we store internally and leaves non-coeluting peaks with a red dot for the transition in the Targets view.

Thanks for your interest in the details. I hope we can make some of them easier to understand in the future.

mvillal1 responded:  2017-08-29
Hello Brendan and Nick, thanks for the replies and sorry for my late reply. I also think it will be useful to have an option indicating how Skyline did the evaluation of peak groups and how it ended up selecting a specific peak group from all the others that can be found in a XIC. However, as of now, even if visually (green/red dots) Skyline is not showing which transitions were considered to be "co-eluting" in other manually chosen peaks... Can I trust the "co-elution count number" values reported in the mProphet features output for all the other peak groups observed in a XIC (i.e., the ones that were not automatically selected)?

Thanks again,
Brendan MacLean responded:  2017-08-29
Yes. Coelution count numbers are correct and separate from what you see for manually integrated peaks. When a user picks a peak manually, we "force" integration of all transitions and only ones with no peak area will show up red, regardless of your "Integrate All" setting. But, internally, we know which transitions were forced and which were not. We just aren't showing you that information. Which we can improve in the future, but for now, your only access to it is the coelution counts exported from File > Export > mProphet Features.

Hope that provides enough clarity.