SomaModules - Pathway Analysis for Scan Output! (And publicly available SomaScan data!)

 

This paper took me some time to dig through, with the majority of the time focused on PUBLICLY AVAILABLE SOMASCAN DATA!

Paper first. It's here! 

What's it about? It's about how to change the output data from SomaScan into something useful. In this case the authors walk you through a large pile of separate R Scripts that will take you from your Somamer output and will get you to a gene identifier for each one....cause...obviously....that's not what you get from the vendor....

Here is what you'll need to take these output data and get to pathway analysis....they all won't fit on the screen....

Now....you absolutely could have all these separate scripts in a single notebook, but it is certainly funnier this way. Transform your silly data format - get this output - put it into this next script - run that - put it into this next one. 

This is a great group and I've got - easily - 500 - Thermo .RAW files from them sitting on around on hard drives here from various studies on the Baltimore Longitudinal Study of Human Aging. 

It would have to be a great group, because they can get usable information out of this SomaLogic data! 

And by THIS SomaScan data - I mean - THERE IS DATA HERE IN VARIOUS FORMS! 

And... it's ...revealing.... I think the paper references it as the 11k data, but I count like 7,400 unique protein targets. So I might have that mixed up, it might all be the 7k aptamer data. 

Just like you've heard, everything appears to always be detected, and it appears to always be detected above the blanks. And if that is your goal, stop reading here and have fun!  You've detected proteins! Probably! 

If you are interested in quantitative differences between condition A and condition B, these aptamers do not appear to magically exceed the physical limits that have long been established for assessing quantitative differences of aptamer - protein binding and release. Nor does complexity of the background and the aptamers present increase the maximum dynamic range in protein quantification ever recorded for an aptamer. Shockingly, it appears to be completely the opposite. 

This data deserves a more thorough analysis. And, as I mentioned, I have hundreds of really good LCMS proteomics file already processed from this study. Most of it is 24 offline fraction TMT data, that I happen to think was done really well. I will, however, leave you with the impressions from my first couple attempts to interrogate these data. 

Deconvoluting proteomics homogenates with single cell type insights??

 

This idea has been proposed to me a few times recently and I thought something like "oh, sure, and exactly how TF would you do it in practice??"

Not sure it's the solution, but I sure need to leave it here so I can look at it later! 

Mapping early blood cell differentiation!

 

Wow, I've been slammed lately! 

Okay, if you haven't seen this, get out from under your rock and check this out. It's stunning. 

More plasma proteomics technology comparisons!

 

I only have a second to drop stuff here so I can close another of the zillion tabs open on my desktop! 

These are going to be intrinsically controversial, so the data we have the better! 

Characterizing and engineering PTMs with high throughput systems (not mass spectrometers!)

 

Disclaimers! I've had this open on my desktop since I got back from vacation and I do not get it. We're way way out of my wheelhouse. 

I do think it's worth posting here, though, for people to think about, if only so I can close the tab and reference it here later. 

Ultimately, I don't think it has any real overlap with what most of us do PTM-wise with mass spectrometrs. We're looking at how all the acetylation sites (for example) change in appearance and abundance when we put drug A into some cells when compared to control cells. 

However, the interest here has to exist here in being able to drive the PTM engineering using in vitro systems. The proof of concept leads to the discovery of mutations that drive up the concentration of the products they want. (Confirmed by targeted LCMS/MS) Still really cool (I think). 

DeCrypt KRAS signaling using all the common inhbiitors!

 

This one had so many steps to get behind the paywall of that I thought I was trying to find a certain global leader's name thousands of times in files about a certain island. 

It was totally worth it though, because this is GREAT! 

KRAS does so many things and has so much redundancy that you can fill entire buildings full of RAS researchers for decades and they still haven't sorted it all out.

Now there are these awesome little small molecule inhibitors, but you're still looking at a big matrix of effects. Tubingen was like - fuck it, we'll just use them all and do super deep 

Proteomics and

phosphoproteomics and 

ubiquityl/ubiquitin/Ubbydooobydoo omics

On cell lines where each particular drug works. Plus 

dose responses! 

Geez. Science signaling is a pain in the butt. Even trying to get the PDF makes you go to a weird PDF reader and then you have to download it from there? I bet the impact factor is lousy....

Got it so I can read it! 

TMT11-plex was uses for most experiments, suggesting they may have started this study before the 16/18-plex came out, or that the Decrypt downstream stuff is just easier with maintaining the exact same workflow - which I'm 100.00% fine with. 

Oh shit - they also did Cysteine profiling! Now....that's cool. I know someone who is convinced these drugs should be having off-target effects. We did some runs to see - didn't - but I didn't have the bandwidth to do Cysteine enrichments.

Now...if I do have a criticism here of this beautiful work I'm not super stoked that some of the experiments were done with SPS MS3 while others were high resolution MS/MS. I'm probably going to download the high resolution runs for looking for things I'm personally interested in and I'm probably not going to download the ones with ion trap MS/MS. 

I'm just a resolution and accuracy snob, particularly when hunting PTMs. 

Don't let that -at all- detract from what a fantastic and useful piece of science this thing is. It's just a minor comment based on how I will personally reuse these data - that you can get at ProteomeXchange here! https://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD063604

Of course, these data are probably already fully integrated into Decrypt so you don't have to download a 300GB of data to dig through yourself. 

However - the last big Decrypt paper that dropped featured some drugs that we've done single cell proteomics on and that data let me vastly improve my analysis of that drug. And - here are 2 more that we've worked on! 

SynchoSep-MS - Another top lab utilizing amazingly complicated HPLC!

 

Ummmmm.....okay.....so when I saw the first preprint with a similar concept, I thought something like "that's a weird idea that probably looks okay on paper but will absolutely never ever work in practice".

And....here is a bunch of top notch scientists like Dan frickin' Polasky, himself, demonstrating a very similar concept? 

Between the 2 studies there are 15 or so authors and I'm pretty sure that means 22 of them are smarter than me - so... 

...wait. That's upside down. 

Do we revisit this idea? It might be like the single ion mass spectrometry thing that I really couldn't figure out, that - to this day - still looks to me like you're just doing scan averaging of incredibly poor signa, but I have to accept that I'm not smart enough to understand why it's something else.

And that's totally fine as long as the science is good! 

Okay - so revist the idea? Here you inject 2 samples on an LCMS system but each one goes to a different nanoLC column so that peptide A for sample 1 and sample 2 come out at different times? 

If I'm anywhere close one getting this - you're operating under the assumption that your MS system is no longer as limiting in the number of ions/peptides/proteins that it can detect as your chromatography is....

Which is very confusing after a decade and a half (geez...longer...? wtf...?) of having way way way more peptides eluting than you can possibly detect - one blog post on the absolutely amazing paper that I was first aware that actually counted this discrepancy. 

If that is true then, this sorta makes sense....? The downsides in my mind that I noted when the similar preprint dropped that you're depending on two nanoLC columns (I'd rather not depend on one - two sounds like 4 times the trouble) still seem valid to me. Also, I'm going to add the concern that this is great for my proteins that exist at 20,000 copies or more, but probably really bad for the proteins in sample A or B that are only about 100 copies per cell.... However, if you're really truly desperate to get that throughput up without mutiplexing reagents, this appears to be an avenue you could explore. 

Let me know how it goes? I like my robust simple little HPLC system where my weak point (my column or emitter) is a single instance. 

MaSSyPup returns with USB - install nothing - data processing!

 (That's a massive puppy!) 

And this is MaSSyPupX (yes, there were some double-takes at the screen at my weekly meetup of dyslexic and dysgraphic scientists). 

This probably sounds silly to a lot of people, but I have friends and collaborators who can't install anything at all on their PCs themselves. When FragPipe or SpectroNaut rolls out yet another super powerful upgrade they have to call (and sometimes pay) someone to come out and upgrade that PC for them. I have control over my PCs for now - but not the HPC - so I have to be careful about mismatching the software versions on each. 

I haven't checked it out yet, but MaSSyPupX appears to allow you to circumvent this completely by allowing you to have an external drive with all your pre-installed and pre-configured software on it. Plug it up and run. Now....even with a lot of external drives being super fast SSD, I expect some latency issues due to USB transfer speeds, but I know those are also improving. Even if it isn't ideal may be a solution to a serious problem or two some people out there have. 

Single cell proteomics defines functional neutrophils sub-types in glioblastoma!

 

Sometimes you just need a vacation from everything. I ambitiously downloaded a couple dozen papers to my kindle but I probably have only read half of them.

I spent some solid time on this new preprint for obvious reasons, though! 

Anyone doing single cell proteomics (SCP) is probably running up against the same problems I am - like....are you sure this is the right technology for the problem? Are you sure it wouldn't be better to figure out a way to get 25 or 500 cells from that sample that are identical that way you can get thousands and thousands of proteins without trying very hard? 

Here we seen an exploration of both large cell numbers (500 cells) and bulk homogenates and single cells. And here is the real kicker - even when we think we have the perfect marker for getting "identical" cells out of that population. 

The coverage here is stupendous for these simple little cells. They may be getting as much as 50% of the total neutrophil proteome on this Astral and - yet again - we start to see important functional heterogeneity in systems that would be a whole lot easier to study if they'd just all do the exact same thing! 

You can do LCMS proteomics on ANYTHING!

 

Not sure if I'm allowed to show this yet - but - w0000000000000000000000000h00000000000000000! 

I'm not kidding - I just put in a proposal for a 1 day workshop where this is the plan.

1) Participants are instructed to show up in the morning with something biological. I'm picturing flowers or pine needles or dirt.

2) We will homogenize that thing - prep with a high temp enzyme (FAST) to peptides

2a) We will find some way of estimating protein abundance fast (work in progress) 

3) We'll load those samples onto EvoTips 

4) At lunch we'll load those samples in my lab (or we'll have someone from my team drive them the 1.4 miles to my lab - work in progress, but the conference is here in Pittsburgh)

5) We'll process those data against whatever genomics data we have

6) Participants leave with a proteomic report of the thing they brought in. 

If you know anyone at ABRF you should totally tell them they should do this workshop. 'Cause this is absolutely where I am today with LCMS proteomics. There is a beautiful vine growing on the dumpster at my apartment complex that the guys who do the landscaping here can't seem to kill and every day I walk by it I think I should take samples of the roots, leaves, stems, flowers and strangely complex fruiting bodies the plant has and do proteomics on it on a Saturday when an instrument isn't in use. 

What I'm rambling about is the fact we did a black widow spider - without a good genome (though we got one later) - and could find multiple points of biological evidence that we have solid functional findings. Since it was a training excercise for a student it wasn't done in a day. But....at the end of his master's I'm pretty sure he could have.... (I tend to make new trainees prove they can do good proteomics on unlimited material before I let them anywhere near things like....patient biopsies...) 

Huge shoutout to Ben Neely for his invaluable help with this project, btw. Dude solves proteomics of understudied organisms in his sleep. 

The Proteomics Show Season 7 - "Alt Proteomics" Podcast update!

Thank you US HUPO for letting us keep rambling into these microphones! I was behind on this, so it's helpful for me. 

This is where we are so far! 

Cecilia Lindskog was AMAZING - (Protein Atlas!) and got us kicked off.

Aleksei filled us in on how computational modeling can be used to build that proteomics instruments of tomorrow (that might end up fitting in our pockets). 

Jeff Nivala talked about how he's already heading that direction TODAY using clever chemistry to yank proteins through off-the-shelf nanopores

And today's episode was the behind the scenes look into how one of the best and most productive O-Link proteomics labs works, with Graham Wiley! 

I forget what's up next but it's funny how much proteomics is out there that doesn't need a mass spectrometer. 

Nova! Build all the mass spec tools!

 

Playing catchup on posts this morning, I guess!

This one looks super useful! 

Don't wanna read? I gotcha. 

https://schweppelab.github.io/Nova/download/

Even more concerns about how cell culture media changes the proteome!

 

During the pandemic I had no choice but to learn /relearn (had done it before, though poorly) how to grow cancer cells in culture. If you've never done it, you should check over the protocols to see what an ...art... it is... Which is what you never want to think about robust scientific workflows.

One of the findings of the best scientific work I've ever done was that the biggest difference between cancer cell lines I was studying with single cell proteomics was culture media additives. One cell line couldn't grow without insulin supplementation and the other could. Ultimately that difference convoluted any direct comparisons between drug effects in those cell lines. 

For even more reasons to be concerned, check out this cool new JPR paper! 

Also -- check out the impressive coverage they get of the HepG2 cells! >8,000 protein groups per condition in 30 min total run to run experiments (TTPro diaPASEF + DIA-NN 2.0). Solid study all around. 

Thermo marketing delivers another tour de force in how to sell million dollar mass spectrometers!

 

Okay - so let's totally focus on the positives. Whoever has been executing these marketing releases from Thermo in conjunction with instrument launches is absolutely undefeated. This is the 3rd or 4th year in a row? Right when Thermo drops a new instrument you get field leaders dropping 

Preprint

Preprint

Preprint (actually, this one is worth revisiting later)

Marketing white paper lightly disguised as a Preprint

Together this adds an absolute shit ton of credibility to every talk you see, even the ones you only went to for the admittedly extremely cool free Legos. 

While I'm clearly being a jerk because this does appear to be a marvelous instrument that will only run you 5- 7  brand new Porsche GT3s

At the end of the day even if the quan at 500SPD looks like complete garbage (I'm not saying it does, I'm saying, if it does). That's still WAY more identifications than you can get from a O-Link or the IlluminaProteinCrap at WAY higher throughput. 2-5x! 

Now - just to make it clear where things stand from a data quality perspective. I would bet my house that an Astral that is operated by a keyboard on the floor of a room full of boston terriers will undoubtedly provide better quantitative proteomic data than the Illumina proteomics solution. O-Link is putting in the work to both improve and prove that it can deliver quantitatively meaningful protein measurements. It's up to Illumina to upend 20+ years of aptamer based protein measurements to demonstrate -for the very first time - that aptamers can deliver meaningful quantitative protein measurements outside of an extremely narrow linear dynamic range. I've been wrong 6 times this morning already, maybe this'll be 7.  I don't think it will, but that would be 8. And I'm fine with that. 

12,000 proteins in human plasma!

 

Do you want to measure 12,000 proteins in human plasma!?!?! 

Is this basically your budget per sample? 

Then do I have the preprint that will finally make your collaborators and customers stop whining about how much SEER costs! 

For every 100 mL of human plasma and 

and 1 week to 1 month of human labor per sample - 

you can stop thinking about what protein enrichment/depletion/fractionation technology to use and just do every single one of them! 

I can see the queue building out the door and down the hallway! 

Shake up your sample prep and a get 2x more phosphopeptides/sample!

 

Honestly, I thought this was going to be boring. How many phosphopeptide enrichment strategies do we actually need. However, when you see how much thought went into this, it is hard to not like it. 

The urea based bead elution is inspired. I wouldn't have thought of that and I'm not sure I understand why it works. However, it is fair to note that the biggest gains they get from this prep is from removing C-18 cleanup prior to injecting on column. Also - and this is really cool - this prep doesn't use phosphatase inhibitors! 

If the absolute number of phosphopeptides in total doesn't blow you away, they are doing Orbitrap-Orbitrap scans and the MS2 scans are done at 30,000 resolution with an auto ion injection time. I think on that means something like 10-15 scans/second on that particular tribrid system. They are clearly biasing this study to quality over quantity. So...2x more high confidence phosphopeptides/run? If you do phospho this is definitely worth checking out.

Oh yeah! And even if you don't do phosphoproteomics the introduction is really worth a read because it highlights why everyone seems so obsessed with these finicky things. 

 

The last post was made because I stumbled onto it while looking for this thing I saw on LinkedIn (I have links from LinkedIn disabled so I don't accidentally end up reading too much marketing stuff). 

Then I went down a rabbit hole watching old Kenny Rogers stuff and now I'm out of time to blog this morning. 

Worth it! 

Solvent swapping on HILIC improves polar metabolite recovery!

 

This belongs on the other blog, but this is the one I have open while traveling so it's going here for me for later!

AND I think one of the authors is basketball and country music legend...

ZenoTOF 7600 performance with NanoLC - and applied to cell-type proteomics!

 

If you've also wondered where a ZenoTOF 7600 would stack up against your hardware when running comparable flowrates (cause you're probably running nanoLC). This new preprint did the heavy lifting for you.

Probably formerly counterintuitive, but something we're all getting used to now - a super fast TOF tends to get more IDs at faster gradients. Narrow those peaks to concentrate that signal and they perform better. As the concentration goes up, that effect tends to level out. 

Based on some vendor marketing material you might be quick to disparage these numbers, but I think we all know that our own labs aren't the perfect operating environment of vendor facilities and after full time use running weird stuff in our instruments they don't run the same at install. 

What is interesting to me is that - yes - NanoLC seems to help, but it doesn't really help that much (maybe 10-20%?) and I really enjoyed the ease of use of the Microflow setup on my ZenoTOF before a ceiling waterfall at Johns Hopkins destroyed the shiny thing. It should probably also be pointed out that this is now a 2 generations old ZenoTOF, as the 7600+ and the 8600 are improved versions of this hardware. Still, a really visually appealing study that answers a lot of questions that I (and maybe you?) had about this instrument.