Jonathon O'Brien

Statistics, Proteomics, Aging, Golgi LLC

[Jonathon O'Brien @ProteOBrien]

I’m very excited to share our new paper on modeling isobaric proteomics data! Thanks to my collaborators @Calico and for some excellent advice from @dschweppe1, @MartinWuhr and @GygiLab. Time for a walkthrough. #TeamMassSpec
https://t.co/IDTV7z372S
1/15

[Jonathon O'Brien @ProteOBrien]

Congrats to Gina Turco @sudogenes for winning the 2024 Parasite Award!
https://t.co/NGSbVTXpeM

Gina is excellent at finding new insights in complex datasets and we are very lucky to have her as a member of our team!

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk But it could have been chance. Lots of the phospho hits were likely at low occupancy. Some of the most interesting divergent peptides didn’t even have an S, T or Y.

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk An all mods targeted assay would definitely make this type of result interesting! The furthest I went was to compare the divergent peptides to a phospho enrichment discovery experiment. It didn’t work very well. A few “hits” might have matched up.

[Jonathon O'Brien @ProteOBrien]

@ProteomicsNews @AmandaLSmythers @1jvaneyk @mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab Once you accept a framework where you check for quantitative consistency, prior to using a protein label, why on earth wouldn’t you apply the same framework to razor peptides?

[Jonathon O'Brien @ProteOBrien]

@ProteomicsNews @AmandaLSmythers @1jvaneyk @mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab No no no. MacCoss isn’t beating up on people who use razor peptides. He’s beating up on everyone! His whole point is that you should expect peptides, unique or not, to exhibit divergent behavior.

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk If I knew how to recommend follow-up experiments, I would probably keep doing this. What sort of recommendations would you make if you found an interesting association in, for example, 2 out of 13 peptides from a protein?

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk There were some problems with the approach. Primarily, my collaborators had no idea what to do with the results. They were accustomed to thinking about gene regulatory networks, or specific kinases. Significant peptides from insignificant proteins were hard to get excited about

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk I'm not sure how to talk to people about different responses from different peptides. If I knew the proteoforms it would be easy, but without that information, it's not clear to me that this is better than the average response from all gene products.

[Jonathon O'Brien @ProteOBrien]

@mjmaccoss @michaelsteidel1 @chrashwood @_AndrewLeduc @DemichevLab @1jvaneyk The gene is still well defined and we can make inferences about associations between the average response and some variables in our experimental design. Similarly, each peptide signal represents ions from some unknown number of unspecified proteoforms.

[Jonathon O'Brien @ProteOBrien]

@NatalieTurnerAu @michaelsteidel1 Also, this is a big issue in the human proteome. I think ~40 percent of human peptides map to more than one Uniprot ID. It's rare the point of being negligible in yeast.

[Jonathon O'Brien @ProteOBrien]

@NatalieTurnerAu @michaelsteidel1 MaxQuant does this. So does the Gygi Lab software. The vast majority of the time, non unique protein assignments map to another protein from the same unique gene.

[Jonathon O'Brien @ProteOBrien]

@michaelsteidel1 @_AndrewLeduc @DemichevLab @mjmaccoss @1jvaneyk I would be interested in seeing examples where this was false. Is anyone making verifiable inferences about proteoforms using bottom up data?

[NimwegenLab @NimwegenLab]

This is not how one does serious science. One cannot show a picture, claim it shows the 'structure in the data' and then be completely unable to explain what this 'structure' supposedly represents.
It's a joke, frankly. n/n

[Jonathon O'Brien @ProteOBrien]

Shared peptides are a problem. At first glance the algorithms look absurd. But the reality of the situation is not as bad it sounds. 5/5

[Jonathon O'Brien @ProteOBrien]

The most common outcome is that all of the proteins in question have 1:1 ratios throughout the experiment which has zero negative impact on your results. It’s only a problem if one of the proteins is doing something “interesting” while the others are not. 2/5

[Jonathon O'Brien @ProteOBrien]

Even when there are substantial contributions from more than one protein, and they are doing something different, having multiple observations might result in the shared peptide being flagged as an outlier, or it might result in a small increase in variance. 4/5

[Jonathon O'Brien @ProteOBrien]

I’ll play devil’s advocate here. Suppose that the observed peptide is actually a mixture of identical peptides from different proteins. How does this usually hurt your experiment? 1/5

[Michael Steidel @michaelsteidel1]

Hi #TeamMassSpec,

Its a proteomics community standard to assign peptides shared between proteins to one of those proteins, even if there is clear evidence for presence for all of these proteins in the sample.

Why do we do so?