Fudging the volcano plot!

 

Picture borrowed from this recipe. But this is what we're normally talking about. 

I feel like I only recently learned how to make volcano plots and I know that there are drawbacks to them. They're also probably overused (like PCA and T-SNE, etc. etc.,) but it does feel like each paper should have some pictures or graphs in them. 

Here's a short discussion of strengths/weaknesses and alternatives! 

Introducing the official mascot of the Human Proteomics Organizations!

I've got a lot going on and I've had to take a step back on a lot of things, including the amount of service that I do. Don't worry, the US HUPO VMO is in great hands with Pratik Jagtap and Jordan Burton taking over as chair and vice chair, respectively. I'll still pitch in as part of the coolest committee in all of the Human Proteomics Organization's respective histories.

The last thing I get to do is announce that the HUMAN PROTEOMICS ORGANIZATIONS now have an offical mascot. I've led this important initiative almost completely on my own. It is the least I could do. 

Introducing - 

THE HUPO HOOPOE! (Specifically the African Hoopoe, or Upupa africana)

(Image above generated by the DeepDreamAI. I pay for a subscription, you can use it, but the ones at the bottom are the official logos!) 

A lot of thought went into selecting this majestic animal to represent the prestigious HUPO and US HUPO organizations, but check out the number of boxes that this one ticks off!

1) It nests in dark holes or caves! Just like most proteomics people stuffed in basements because once-upon-a-time mass spectrometers needed big magnets and that made sense. And now it's convenient for real scientists who would rather prefer that proteins (and especially mass spectrometrists) aren't a thing! 

2) When they are out of their caves they make a lot of noise! You can listen to HUPO HOOPAE sounds at this YouTube video!

3) The Hoopoe have majestic crowns upon their heads, which....hmmm....it can't match every characteristic. 

4) All animals (and even bacteria) avoid these, not just because they're saying "transcript numbers have no relationship whatsoever to the amount of protein around" over and over again, but because they smell very very bad. 

5) Perhaps most importantly, things don't bug HOOPAE, particularly not within their caves. They have evolved not only the unique passive defense mechanism mentioned in 4, but they also have unique active defense mechanisms! More details on those here! 

Here are some of the new official HUPO logos. These will also be available on the official US HUPO and HUPO websites soon, but you can download them directly here. 100% reusable art. I actually generated the vector image myself with no AI. 

Admittedly the "action logos" have been somewhat more controversial, but evolution is responsible for the development of our mascot's unique defense mechanism! I just pointed out the obvious. 

I'll put up a folder later. I'm currently just adding the new art as I generate it. 

Microsoft Designer is going all out to help!

Fragment ion intensity prediction boosts TIMSTOF peptide ID rates!

 

This was on here as a preprint, I think, but I also think it got better during peer review.

The idea is pretty simple. The thing about the millions of spectra that things like PROSIT use for deep learning is that they were generated on Orbitraps. Makes sense. Super high quality spectra. However, fragmentation energies and energy ramps and mass analyzer architectures are all different between this and other instruments. Heck, it's probably not out of the question to think that the Orbitrap-TOF Asstral fragmentation patterns are probably not perfectly matched to Orbitrap spectra. Maybe they are, but there are definitely differences between other vendor TOFs and these library spectra.

This group works through those differences and finds they can drastically (3-fold more!) improve identification rates by retraining the models! 

US HUPO Early Career Researcher Seminars May 17th!

 

Need some valuable insight from a panel of researchers from across the range of academic career stages? 

BOOM. Here ya go! 

https://us02web.zoom.us/webinar/register/WN_RNWI3ua4TN-jhz65DJJQuw#/registration

De Novo Multi-Omics Pathway Analysis outperforms gene based pathway data!

 

I can't follow the maths in this paper, but I appreciate the results! 

Most of the stuff we use for "pathway analysis" is based on historic data from....somewhere.... and the details of that somewhere can have some effects on your results. That ...somewhere.... is often data from the past like RNA Microarrays or yeast 2 hybrids or things that have become very niche things because they were replaced with technologies that are more effective for most applications. 

Since things like Ingenuity Pathway Analysis are drawing from thousands of RNA Microarrays - nearly all of which were used for studies of cancer - those tools often aren't all that helpful for things that are not cancer. Do we need these historic datasets? Could we do better by just throwing in our own multi-omics data? 

This group sure thinks so, and the results are very encouraging.

But you don't have to believe them. Neither do I. DMPA has been compiled as an Executable program (or you can run it in something called MuttLab? Woof). and you can get it here! 

JHU Mass Spectrometry day is back! With remote options!

 Thought I posted this a few days ago and couldn't find it. 

Interested? You can register here!  

DEEP LEARNING DEATHMATCH - What actually predicts peptide fragmentation best?

There are all sorts of thing out there now that can learn from millions (billions?) of MS/MS spectra and can take your sequence and predict how it will fragment.

Which one is best? Who knows? Maybe we should have an algorithm DEATH MATCH! 

Okay, again I'm scrambling to wrap up things before they take my keys and kick me out on the street, so I'm leaving this here mostly for you (well...me...maybe...cause I do want to be able to find this later).

HOWEVER their analysis turns out this tables 1 and table 2 are ridiculously valuable. 

They're summaries of immunopeptidomics datasets! Which is what they tested prediction capabilities against. 

Profile rare cell populations with the sacrifice of 1 (one) lab animal!

 

If you've never done it, it absolutely sucks to kill a lab animal to get samples. Not only is it awful but an increasing body of evidence suggests that those millions of years or evolution make it not make very much sense at all to do it for many mechanisms. However, there are some systems where you absolutely have to do it to get that information. What if instead of needing to murder dozens of animals you could learn everything you wanted from just one? 

Hey, wait, this is one of the few labs on earth who are doing real single cell proteomics with mass spectrometers? What are they doing using more than one cell? 

Answer: I suspect having a much better day with their 500 cells than they had with most of their days of working with one! 

Besides being easier, this makes a lot of sense, right? Outside of marketing literature and some people studying single cells that are larger than my dogs, we aren't getting what we expect in terms of today's proteomic depth. But you can FACs sort 500 cells (assuming you have good cell markers) from just about anything.

This group goes through and quantifies something north of 7,000 proteins from every known cell type in rare c-kit+ progenitor population! (review on that here, I didn't know what it was either). 

Now, a fair argument that you could draw from reading other work from this group is that we might not actually know every cell type that is there, but at 500 cell resolution and as good as they appear to be at FACs you can learn a ton. 

While saving animals is a super worthy endeavor and some biologists might be like - ummm....what about animal to animal variation.... - it's one hell of a proof of concept. Some of those things where we can't possibly do human research is because taking a big chunk out of someone is generally not good for them. But you can lose 500 cells from just about anywhere without it being a big deal. Pretty easy to start imaging the future of low input proteomic diagnostics, right? 

Stellar preprint. Highly recommend you spend more time on it than I did this morning! 

Got last generation's instruments? Here's how you optimize them for ultra-low concentration samples!

 

There it is! It was ASAP when I read it on my phone the other day and it moved to this month's issue. 

I might be getting old because the HF-X and Lumos both seem pretty recent to me, but in a review we're putting together we refer to them as "previous generation instruments" and that does appear to be the case.

So...what if that's what you have and someone wants you to run 1 nanogram of peptides or less? Do you want an okay number of peptides and proteins? Or would you rather have 15-fold MORE? Probably the latter, but you do you, yo. 

The reason I took screenshots of the paper with my phone, however, was where and how the peptide "supercharging agents" (as you'll see them referred to in some other studies). DMSO and NBA are employed here to improve reproducibility. I'm pressed for time, but if you're interested in low concentration sample optimization there are a lot of gems in this study. 

Extracted ion chromatograms to improve peptide ID and quan!

 

Now, this study is very clear that this workflow is simple for me or you to add into your or my workflow by just following 3 steps in the materials and methods section and Imma take that as face value. 

because I really like the conclusions. Such as -

and 

as well as 

So when we get through the very last set of mass spectrometry experiments that I will ever run at the Johns Hopkins University (on the systems right now! eeeeeeeeeek! cross your fingers for us that we can get just a couple more weeks without another flood) I'd really like to try this out.