ADAPT-MS - A starting point for automatically classifying clinical (untargeted) proteomics data!

This one took me a couple of rounds of putting it down and coming back to it later.

It's a smart concept and a very nice thing to think about as proteomics becomes more trusted as a diagnostic. 

I think I first thought it was something that it isn't, and that's why I had such a conceptual problem with it. Obviously, I might still have it wrong, but this is how I'd describe it. 

What if you had a random patient come in and you could do global untargeted plasma proteomics on their sample? Not inside of a controlled cohort that you planned 2 years ago and pulled all the samples from the repository? Just that one sample that just came in. That's how clinical stuff might work. A sleepy 22 year old might be working nights to save for grad or med school and be studying and run those 12 samples that came into the lab (typically because it's super important) at 3am. Could you do anything with global data? 

If the answer is no, then the future is not very bright for diagnostic untargeted proteomics. If the answer is shmaybe, then you're getting somewhere, and if it's a yes, then let's start building on this idea right this second.

To simulate it they pulled some traditional proteomic studies where they had a discovery cohort and then a validation cohort and someone did it all the traditional way. Found the markers in batch 1 and focused on how well that marker seemed to be predictive in batch 2. So these authors loaded those data, pretended they didn't know what went where and use the machine learning things to try and sort it out - and it totally ends up doing okay! 

We've got ourselves a shmaybe here! 

I appreciate the transparency of the authors, the conclusions almost read like a "limitations" section. The rest of the paper reads like someone was sending a secret code to Olga Vitek that only she would be able to decipher. If that was really what this was, Nature page fees may be the absolute most expensive way to do it....

Here is the thing, though, it didn't outperform the traditional human thing when the experiment is done really well (the example data they used is superb, probably outliers) but it did reasonably well, and that's still a huge deal. 

 And everything to reproduce it yourself is reasonably well annotated in these notebooks. 

Multi-technology analysis of human liver diseases!

 

I'm tired of reading today, but I really want to get back to this cool paper.

Really deep multi-proteomic type analysis looking for markers for why almost everyone has liver inflammation, but for some people it's a really bad thing that progresses to worse things.

You've got secretion proteomics, and neat plasma and depleted plasma, and some SomaScan data from a related study that they used, and normalized, but don't go into much. They describe the statistics and provide the output data as an excel spreadsheet, which I very very much appreciate. Really nice super high speed targeted work (5 minute gradients on a SCIEX) and just a whole pile of really cool stuff to dig through! 

EAciD optimization for glycopeptides on a ZenoTOF!

 

Oh. This is really cool. I'm so glad that SCIEX is finally getting some traction with their super cool ZenoTOF hardware. The high speed high resolution mass spec world is really super competitive right now. You really can't make a bad choice (aside from Agilent, obviously - hey, I didn't tell them to abandon global proteomics instrument market -they're doing fine in their chosen niches) for getting amazing proteomics data.

There is exactly one instrument out there that has big ass magnets inside it that forces charges onto your peptides for democratic fragmentation. I was supposed to evaluate a ZenoTOF for single cell proteomics, and it was good enough to get a paper in just a couple of months with it but the super sensitive fast PRMs (which got us a super cool paper in the same time period) and EAD were what really wow'ed us. 

Could you do even more with EAD, though? What if while you had those ions you also applied another collision energy? Could you really bust those molecules up to get complete coverage of the peskiest ones? Importantly, could it still be way faster than other democratic fragmentation methods (that use chemical based fragmentation)? 

Pretty much! Best I can tell they optimize these glycopeptides out and they do need to slow it down some. It looks like the best data is coming off between 13 Hz and 19 Hz (my math from the method section details). They have some time for the EAD and some time for the CID and some accumulation time and that sums up.

I don't know what the new Exedrin ETD benchtop instruments are getting with their improved (and seemingly incredible) new Orbitrap hardware. Given I'm used to 100ms reaction times for ETD not counting the Orbitrap scan times, and internal HCD cells (IRMs?) stop, gate, go times. I think this has to still wildly competitive. 

Probably also worth considering that the 7600 this was tested on is now 2 generations behind the faster and more sensitive ones. So...I suspect you could go even faster on the new ones? 

Histology guided lipidomics and proteomics with co-registration of spatial information!

 

Are you ready for deep visual lipidomics? No? Okay what about normal spatial lipidomics with deep visual proteomics? 

The diagram pretty much shows what the paper did, but the co-registration of the spatial data from the laser capture microdissection work (now called deep visual proteomics, y'all, get on the hip terminology) to the lipidomics is a star. This does come from the very small but valuable bank of post-mortem human brain tissue in Baltimore that one of these authors definitely didn't steal from another facility. 

Lipidomics was at 50um resolution on a Bruker TOF I'm not familiar with, but they did both MS1 and MS/MS analysis. The deep visual proteomics was done on approximately 500 cell cuts. Pretty cool since we learned yesterday that brain cells are small. That's 25nanogram or so? Small! 

TMTPro was used for the proteomics with MS2 on an Orbitrap Fusion II (Lumos) system and found around 300 differential proteins of interest that the authors seem interested in. >40k peptides are reported, which is pretty darned good from sections this small on this hardware. If you're thinking of taking the spatial proteomics plunge this seems like a great resource for taking that step. 

Single cell proteomics of the developing human brain!

 

Big thanks to Matt MacDonald for sending this last night with a "Wow" as the total email content so that I got to sleep at like 1am after I felt like I'd finally gotten through it. 

Honestly, "wow" is still the correct word for it. 

Before I get into it, this is the paper. 

Single cell proteomics still feels new, but maybe I'm just old, but we're still learning what assumptions we need to make to get to real biological discovery. 

Something I argued for years was that I'd much rather have more cells than more coverage, but I think I've fallen headlong into the coverage race along with everyone else.... this paper is a solid smack in the face because they did A LOT with a few hundred proteins per cell. 

They say they get 800 on average in most of the cells. I'm spot checking in DIA-NN and before bioinformagic, I'm getting 450 or so. Probably by the time you match between runs and stuff you probably can double that. I ain't reprocessing 1,500 files, so I have a clear sample bias. 

Edit after this post blew up - I kept forgetting to mention the size of the cells, which is a big deal. They think some of these are like 50 picograms of protein! These are like 1/3 the size of the cells we use as our control cell line in my lab. This is a big deal. 

And - this is going to sound critical - and I don't mean it to be that way, because this is just a stunning work - but mass spec proteomics people may really just care far too much about quantitative accuracy. This isn't the first time one of our key tenets of proteomics has been really challenged. A Slavov lab study made the heretical decision of not fully resolving TMT reporter ions at baseline. Something that has been unthinkable for a decade or more. It still totally worked. We may try it ourselves sometime here. 

This study did around 2,000 single cells, about 1,500 of them from "brain cell types" by cranking their resolution and ion injection time to the moon on an Orbitrap Fusion III (Eclipse) with 40SPD "Whisper" on an EvoSep (the 100nL/min one)

I'm not looking at the paper now, but my notes say that it was DIA with 50Da windows and 250(!!) milliseconds of fill time at 120,000 resolution per MS/MS event. With 12(!!) windows.

12 x .25s x 1 MS1 (which might have been 240,000 resolution) so 3.25 SECONDS? per cycle. Someone somewhere in Seattle was shown this line I just typed - and threw up. All over the place. 

But hear me out. For real, what if your quan doesn't have to be good? 

Whew! Files finally downloaded so I could look at some of them and --- yeah -- you're going to get a lot of peptides, maybe most of them, with 3 scans/peak...

One of a pile of peptides I've pulled out, but look, I'm a blogger and I have a lecture due today for a class I'm teaching next week and 1-2 feet of snow between me and work, download 'em yourself here if you don't believe me - 

ftp://ftp.pride.ebi.ac.uk/pride/data/archive/2025/12/PXD071075

Here is the point to stick output of the peptide above. It looks like a triangle, but it isn't actually as good as a triangle would be. 

Is the area under the curve of this peak a reasonable approximation of the signal of the peptide? Who knows? Not me, and not these authors. But is it probably reflective of whether one of the 800 proteins in this cell is higher than the 800 proteins in another cell? Probably! At this depth you're going to be doing a lot of presence/absence stuff. And in this model that is probably a lot of power! 

OMG, and I have laughed for real multiple times about S.Table 2. Man, did they throw some shade at just about every label free single cell study that did fewer cells than this one did! Wow. I would like to thank these authors for not citing me, therefore I did not appear on their Table of Shame. They wouldn't want my stuff there because with SCoPE-MS/SCoPE2 this is actually a very normal number of cells analyzed, but the authors made it very clear that label free quan, regardless of how poor, is the superior option. They might be right. They certainly convinced Nature Biotechnology to accept $12,300  to convert this work to PDF and post it on their website. And if that isn't evidence of a good study in 2026, I don't know what is. 

Okay, but the take aways here should be 

1) You can do a lot with a lot of cells 

Even if!

2) You only get a few hundred proteins per cell

3) And the proteins you detect aren't all that well quantified! 

A good experimental design and cool samples and solid informatics can push you through to an amazing study. 

Quick math, btw, at 40SPD these 2,300 runs or whatever ran - with no blanks, no QCs, and not stopping to calibrate and no failed cells (there are always failed cells) a little more than 2 months on a system that is a couple generations back. That's...not bad....

And they used FACs so the cell prep was inexpensive. I don't have our calculator in front of me, but I'm going to go with this being in the $20/cell range in total costs/cell before any labor. Possibly less. 

Kid's up, gotta run. Super super super cool paper you should check out! 

MALDI imaging on a super cheap little benchtop TOF?

 

I want to close this tab on my desktop so I can see other tabs. 

Direct link to the PDF. 

For real, I think this box is cheap. Like I've seen it second hand for less than an HPLC. Even if you needed to buy some expensive reagents, it could be a legit way to get yourself into the world of MS imaging...

Compare mutliple methods of gas phase fractionation on TIMSTOFs for neat plasma proteomics!

 

Backtracking this post, not sure how I forgot to put it here! 

Frag N' Flow! Fully optimized FragPipe for HPCs!

 

Another one I missed and am coming back to so I don't lose it and can finally close a tab or two! Figure B above at the very least could save you a ton of time over optimizing this yourself! Insanely useful. 

CRITICAL REMINDER - FragPipe is free for Academic use only! 

If you're not an academic, please play by the rules and get a license from Fragmatics!

If you're an academic, I think you could use this whole thing!

 

Single cell SDS-PAGE!

 

Wait. SDS-PAGE has enough sensitivity for single cell proteomics?  Hmmmm.... It probably does....at least for high concentration band recognition...and it would provide some level of proteoform resolution. 

I'll be honest, I first thought "...that sounds slow and silly..." but the more I think about it the more I like having this around as a concept - or even a first pass.

They use the migration patterns for their output and then use statistics that can tell different single cancer cells apart by those patterns....

I don't know about the 3D imaging part, and it is a preprint, so grain of salt over your shoulder or whatever, but I'm definitely going to think about this on my commute. 

Single arabidopsis cell proteomics!

 

The preprint is also on this blog somewhere, but it's cool to see it out! 

Proteomic (and transcriptomic) map of 28 primary cell types!

 

Great new dataset alert! 

28 different primary cell types! What a treasure trove (is that a word? I feel like it's a word. Like valuable stuff you'd need to sort through?) 

Primary, in most cases means something like "we didn't get this from a cancer patient in 1958 and somehow it is still growing and mutating a century later". It can mean different things in different contexts, though. Sometimes it's cells that won't divide, but they will stick to plates and divide for a little while. Just bringing this up so you're cautious about use of the term around biologists and pathologists.

It does give the feel of maybe a pre-pandemic study that finally got prioritized for writing. Orbitrap Fusion 2 system, DDA, SCX or SAX fractionation involved. That doesn't mean it's bad, by any means. It means it's high resolution fractionated DDA data that took way more time to generate than if we ran it today on one of the fast DIA boxes. In fact, it means a dataset that could yield new findings in the future, you'll just have a lot more (and smaller) files to keep organized. 

PTM analysis was done using BOLT! Yeah! First paper I've seen with this cloud based search engine (that I'm very biased about due to like 5 papers I'm on about it, including the very first one) for a while. If you're not into advanced PTM analysis and worried about that, the data was also analyzed with MaxQuant and the results summaries are available in the Supplemental and on PRIDE as PXD062642. 

Global proteomic structural analysis IN LIVING CELLS!

 

Dropping this here so I can get back to it! 

The omics molecule extractor! What a fun and easy way to visualize aptamer data!

 

All, this is legitimately a very nice and very easy to use tool.

You can load metabolomics data or transcriptomics or even proteomics data into it! The test data is some previously published aptamer data, and it's really cool to look at. 

You should check it out. 

You can go right to the online tool here and start pushing buttons. 

I haven't analyzed a load of SomaScan data before, but I've got a couple datasets and the Omics extractor can simplify my pipelines for looking at it down to a single push button. 

Check this out! Lots of proteins detected at just about exactly the same amount in every single sample! In this case it is people with or without arthritis! Look at that precision!  If I didn't know better I'd think that aptamers bind to and offload from proteins but do not do so in a quantitative manner outside of an extremely narrow dynamic range. 

Cricket enriched pasta proteomics!

 

Google said I could use this image, and it's amazing. 

Seriously, though, this is new study is also really cool. 

Look, I have no idea if the deoxidation potential of crickets as a food additive has any scientific merit. I can't possibly know this, and legitimately have no interest. I am, however, aware that the global population is still expanding and the climate is collapsing and no one is going to do anything to stop either thing. Other food supplies are going to be necessary possibly within my lifetime? Definitely within my child's. And we're going to need to think about the allergenic implications of doing things like introducing 20% ground crickets into our spaghetti.

This paper focuses on the benefits of adding cricket protein, but - wow - do they do some cool stuff with some endogenous peptides in ultra-complex matrices. Just about what you'd do to look for allergenic peptides. The LCMS was an EvoSep and a TIMSTOF Flex. Maybe the inflammatory stuff is real as well. 

Super fast targeted proteomics on an ion trap for IBD monitoring!

 

Sometimes you already know your targets, but porting them to a targeted assay is BORING, and typically requires one of those crappy triple quads. One transition per peptide? Repetitive and boring and you basically get a true/false for each transition per scan. Is it real? Is it coeluting in this one patient? Good luck sorting that out! 

Could you use a super fast, crazy sensitive, but absolutely overpriced (come on, y'all, you've got the only ion trap, you could claim targeted proteomics market share) ion trap to get PRM (multiple transitions per target!) to biomarker studies? 

Maybe! Wait. Maybe if someone proved it could work to the QQQ people then you could maybe then try to corner the targeted market? 

This group went up to 300 SPD with almost 1,000 samples! 300SPD wasn't everything they hoped and dreamed, but the quan at 140 SPD looks great! That's still a 1,000 cohort study in a couple of days. Okay, it's several days, but still. I'm impressed. Now...if only the clinics could afford it..... 

Deplete out 99% of the dead cells to clean up your proteomic data!

 

Even if you're not doing single cell type studies, chances are you've got some percentage of your data that is being biased by nonviable / dead cells that may not represent what you're trying to study. We kept trying to do single cell on toxicity models and - if you're doing an IC50 study...50% of your cells are dead. So....do you care about the dead cells? And do the ones that are not dead, or possibly more resistant, represent your phenotype accurately?

What if you could easily deplete out the dead ones? Would that clean everything up? 

Sure looks like it in this case! Over a 55% increase in signal for the proteins this group cares about. AND they got a list of high abundance proteins that seem to drop off when they remove dead cells. New markers for excessive cell death? Sounds like that to me! 

How does COVID affect hamsters? Proteomics answers the big questions!

 

I had a short day last Friday because one of the kids that plays Paw Patrol with my kid had a positive Covid test. Wait. Is that still a thing? Yes it is. So the very first thing I wondered was "I wonder what the virus does to hamsters???"

Proteomics to the rescue! 

If you're not good at reading or a conservative or both, the two seem to correspond, I will clarify that I'm being facetious here. 

I legitimately think that some friends and I were among the first people in the US to have Covid. There was an international trip and one person had just flown in a very long way east from a meeting and we were so so so sick. But there weren't tests then, and who knows?  And who knows what other people fall in that category of "possibly had the virus, but we'll never actually know?" How do you do "changed by virus" studies on people who may have had it? Or may not? 

Apparently these poor little rodents express all the correct proteins to make them a good model of something that can get the virus, but probably hasn't. So the animal testing is justified by the fact it's probably impossible to build a good human cohort that hasn't been exposed. I hate animals models. Abhor. Loathe. But I also can't CRISPR the most abundant protein in a human's hippocampus either. 

Animal model...is...I'll grudgingly admit, the only way to do some things. 

Proteomics was done by DIA on an Orbitrap III (Eclipse), pooled samples were used to make a spectral library then DIA was used for quan. Phosphoproteomics was also performed and a pile of IHC and validation was also performed. Solid looking study. Easy to write a silly headline. A+. 

Proteomics of butterfly metamorphosis!

 

A little disappointed by this picture, but nothing else in this great new study! 

Fixed it! Wrong butterfly, but I don't care and neither do you! This is the whole genomics vs proteomics argument - 

same genome - different proteome! Where are the wing proteins in the early life cycle stages, Mr.Genome person? Checkmate. 

Let's go! 

The data was acquired on a Q Exactive Classic using nanoLC and DDA. Data was searched in Proteome Discoverer 1.4 (I guess they could have cited something for the software. I wonder where they'd find a reference for that?) and a RefSeq genome with 19k protein sequences (whoa!) was used for analysis. Okay, peptide mass tolerance is too high by about 20x for this instrument, but maybe they did that to increase the number of bad hits for the FDR. The method for estimating FDR isn't disclosed, so it might be anyone's guess. 

Oh. Weird. I think they spectral counted. In 2025 2026 on a Q Exactive which has onboard electronics to limit repetitive peptide sampling...? Interesting.... And they used a microarray normalization R package for differential quan....which...is.... whatever. 

They did pull data from a previous paper where they did transcriptomics on another pile of these butterflies, which is cool, and demonstrates that basically transcript abundance is a pretty uselessthing to do with your time. 

The data is up on ProteomeXchange if you wanted to do a reanalysis with a less eccentric data analysis pipeline as well! 

New ScienV biosketch format - ORCID won't assign? Possible reason/fix.

Happy 2026 US Researchers! 

Here is your friendly reminder that the NIH did have time to invent not one - but 2 (two!) new required Biosketch formats. Ignore the one from the summer. Starting January 2026 your BioSketch MUST be made through the ScienV portal at the National Library of Medicine. (Yes that still exists, it's just the Goddard Library that is being shut down and the books thrown out).

I assume the NIH came back from the longest shutdown in US government history as surprised by this as we are, so there is a glitch or two. 

If your ORCID won't connect (which will completely prevent you from saving and printing a BioSketch) you can check it this way.

Log completely out of your internet browser or use "IgconoYolo mode" 

...you'll find that in the upper right corner of your Doogle Chrome AI enabled browser....

Make a note on your blog to investigate your 28 compromised passwords and wonder if that has anything to do with the alert you received from Chase yesterday of them rejecting an attempted $228,000 purchase. Definitely investigate. I figured the Chase thing was Phishing...?

But THEN go to the National Library of Medicine https://account.ncbi.nlm.nih.gov/

And try logging in through your ORCID. In my case I found a totally new account registered to -

...which....is neither my email...nor a valid email format. I don't have a lot of publications or grants, but this NLM account was linked to zero. And I have more than zero.

I reported this to the capital letters of National Library Medicine dash support@nlm.nih.gov - which I'm sure they're very happy to have listed publicly on an open to access blog. 

Boom! Now my fictitious NLM account is gone and it can be registered to my ScienV portal -I think. 

A new pile of mass spectrometry patents!

 

Link to the original post (with patent links here). I tried opening it in a browser I'm not logged in to (into...? en to? e tu? something) and it worked! 

Add another ion funnel for absolute sensitivity!

 

This is probably a great idea, but it ABSOLUTELY has a funny parallel I haven't been able to stop thinking about.

Wow. Did I ever go down a rabbit hole. I was looking for a MadTV skit and then discovered that SNL did it first - in 1976! SNL did a 3 blade razor. 

MadTV stepped it up a little with 20 ion funnels! razor blades! (Youtube link)