TesorAI Chat - An AI (genAI) designed for downstream proteomics data generation!

 

TesorAI has been on the blog somewhere before. As an academic I very much approve of the fact I can now process up to 500 Thermo RAW files/month! 

Edit/corrections - when I first starting using TesorAI it was only Thermo RAW file compatible. I'm punishing it right now and this is clearly no longer the case. 

The link above will take you to a high speed 45 second video demonstrating what you can now do with TesorAI processed data. You can tell a freaking chat bot to make your data into heatmaps and then just tell it to change the colors or add a heatmap - AND IT WILL DO IT. So so so impressed. Now....how does it do with 2 million human mutations as input sequences? 

Check it all out here. https://www.tesorai.com/

moPepGen - Find noncanonical peptides with graph based algorithms!

 

This new tool may only be exciting for a small group of people - but holy cow, yo - this is a just visually stunning. Maybe I'm just used to the fact that all LCMS proteomics figures generally come off the software looking like crap, but graph based algorithms seem like a great idea. 

You will need variant input sequences from transcript level data. And proteomic data. And someone who can use python - and can install a docker. All that magic stuff is here. 

Regeneron and Geisinger announce 200,000 patient proteomics study!

 

Watch out UK Biobank - Pennsylvania is coming for you with 2 HUGE proteomics initiatives!

Geisinger Health is well known for monitoring the genomics of my ...neighbors who prefer to use horse drawn carriages over motorized vehicles. Due to the limited speed of ...horses....people don't end up too far from home so it's easy to track geneology over time. Geisinger does other things, but that's what they're well known for. And Regeneron has the deep proteomics know how to pull off a 200,000 patient cohort!! Press release is here. 

Big DIA software advances - peptidomics and PTMs are the clear focus!

 I just upgraded to the fancy new SpectroNaut 20 and was flipping through all the things! 

Here are the release notes (in screenshot form, which is by far, the most convenient way to share text!) 

And I'm having the guys at the HPC upgrade FragPipe Academic and DIA-NN Academic to the newest ones for me as well (I don't have permission) and in the latter case (DIA-NN Academic) there are a lot of shared concepts!

In case it needs to be typed, I'm not accusing anyone of sharing notes. These teams are very user centric and we are all probably complaining about the exact same things.

Why can I find more PTMs in my non-enriched DDA data? 

Why is DDA so much better for peptidomics? 

Why isn't this faster? 

And here we are, at a great time to test some cool stuff out! 

Symposium on proteomics in the life sciences at the Broad! Registration open!

 

Wow. Sort of short notice, but this looks like an amazing use of a few days in Boston/Cambridge. This wasn't on my radar for the summer. The program is a little vague, but the organizers are some of the best scientists in proteomics and have picked extremely relevant topics. Definitely something to think about! 

Edit: I found the list of speakers as well. I was just on the wrong page. Geez. Check it out, the week is stacked. Here are the main themes. 

TMT in FragPipe with TMTIntegrator!

 I gotta move fast, but for those of you on the TMT32plex or 35plex or whatever it is, I think you'll really like this screenshot! 

How? What? I don't know! Check it out here! 

Peptides bistability in ion mobility might explain a lot of things!

 

Okay - somewhere on this blog a few years ago is a post that is called something like "what the hell is ion mobility"? And I'm continually learning - or trying to - and might have just given 3 talks doing proteomics with ion mobility devices. I might still only partially know what it is and I certainly couldn't build you one.

However, I'm occasionally surprised by measured versus predicted ion mobilities. And I'm often troubled by how ion mobility clouds are altered by things like isobaric tags. And for the former problem....this new preprint is really really interesting.

Why couldn't a peptide have multiple ways to be pulled through a gas gradient? I mean...it seems like you could probably flip 50% of your peptide population A one direction and 50% the other and they'd have different "shapes" - and it's not like the peptide bond is a crystal lattice. (Lettuce? Now I'm hungry). It seems like a curiosity until you see how much their IMS predictions improve when the model allows the entry of 2 stable IMS readings for each peptide....

ASMS2025! Press report 1 - Proteomics Hardware Advances!

 

Guess who finally got the ASMS PRESS PASS! Now....would....it have been an easier conference to have given 3 separate talks and not had a bunch of press responsibilities? Yes. it would have been. 

Waters saved me a bunch of time by inviting me to their conference but then not letting me in, so I used that to sit in my room for just a few and stare at the wall in the dark to recenter, and I sincerely appreciated that time. I don't think they had anything in proteomics to show off. I just went to be polite.  

On the proteomics side the big 3 had big hardware drops.

SCIEX dropped the 8600 which could possibly maybe put them up there as having the most sensitive high resolution system in the world? Maybe? Those numbers are crazy. Let's see where they go. And the proprietary quad sliding stuff on top of it makes it a really interesting and competitive bit of hardware. This launch was, of course, overshadowed to some degree by a very public disagreement with a popular open software development team. I'm not paying attention, but people being open minded about demo'ing different big proteomics solutions seemed to be paying very close attention to it. GenomeWeb dropping a story on it during the conference couldn't have been the coolest thing for their marketing people to deal with. 

Thermo had the Astral 2 Zoom. Higher sensitivity, 270 Hz max speed (up from 200), capable of doing TMT 32-plex -with some important caveats. Like the TIMSTOFs you have to do 2 scans for TMT. One for the peptide sequencing and one for the low mass fragment ions. In the second scan the fragment ions spend more time in the TOF region so they make more passes and then they get the minimum resolution to separate out the ions. The bonus is you can optimize the high mass and low mass fragment acquisitions - and - of course, Proteome Discoverer first launched when the first Orbitrap XL was out there and it had to do 2 MS/MS for every TMT/iTRAQ, so it can take these data. The Astral is super fast, so even 2 scans/peptide and even when one scan is a little slower, is probably really fast.

Thermo also had the Excedrin? Excedion? Benchtop high resolution with higher scan speeds. You can run your Orbitrap at 3,000 resolution and get 70 scans/second. It also has ETD built in. Who would want 3,000 resolution and ETD? Anyone worried about the collapse of their transient in the Orbitrap! Big intact protein, people, for sure! The signal optimization seemed to allow bottom up where the MS/MS actually looks pretty good at 3,000 resolution - from a signal perspective. Some people have noted the mass accuracy seems a bit wonky. Not something I'd worry about, Makarov probably has that fixed already, but it's not like you can't run a higher resolution on it. 

EvoSep had the EvoSep 2 (Eno). Higher pressure loading pumps, better chromatography and higher signal. Looked very plug in play for anyone with and EvoSep One.

Bruker had a pile of launches as detailed in earlier blog posts. The TIMSOmni seemed to be stealing the show on the proteomic side, but we also saw an improvement on the TIMSTOF Ultra2 in the addition of the Athena Ion Processor. (Box now says TT Ultra2 AIP). From van Eyk lab stuff, it looks like a solid signal boost but it also adds some interesting level of control to the Ultra2 that seems under-explored. Imma send some samples up. The fun thing about the TIMSOmni is that you just turn off the OmniTrap and you've got a TT Ultra2 AIP. 

Man - was there some cool automation stuff - everywhere! Robots for proteomics!! And downstream data analysis. I'm currently demo'ing Mass Dynamics stuff and I'll post how it goes here. 

OH. And this was a big one for me. New vendors for N2 generators!! Parker was there and excited to talk to anyone who was tired of the generator that came with their instruments but Swissgas was a completely new vendor to me and we talked a lot about longterm cost breakdowns. It's one thing for me to whine about my N2 generator. It's nice to have useful information to share with people when they say "okay, what are my other options?" 

Alternative suppliers for some things you use a lot? That probably shouldn't go here, but if you were there, you could see where maybe someone else was like "we're also very good at making stage tips..." or "would you like to try one of these things?" I'm rarely going to turn down free stuff even if it is a stranger offering me something that looks suspiciously like a captivespray emitter. I'm probably not going to try it out, though. 

I was part of the Mobilion BILLIE launch (and, again, I'm a compensated member of advisory board) and that was really fun. Hopefully data will be up on PRIDE super soon so you can look at it. All of the data showed was on Agilent QTOFs that were not designed for proteomics, so it's super impressive to see the instrument doing competitive proteomics. It was super cool to work with that team and to field questions like "what could that front end do to a system that was already a good proteomics system??" And I'd also like to know that answer! It's currently a completely vendor neutral system, so hopefully we'll all know sometime soon.

Now, more than ever, proteomics needs better chromatography! ASMS 2025 edition!

 

As my very first takeaway from ASMS2025, I'd like to start with this paper that is just a little bit older than I thought it was when it came up in my talk on Monday. 

Oh man, did we ever see some awful chromatography this week. Whole proteomics on 5cm columns was often higher resolution LC separations. There are 4cm columns and probably shorter. And often at low pressure. Now, maybe all the hardware advances that have happened on the gas phase should negate the statements in this commentary? But I'm skeptical that is truly the case. And maybe I'm just old and old  fashioned, but proteomics did seem to be the one place where the LC seemed to be left the most behind. I popped in on some food chemistry talks and a couple environmental and - they also have big complex matrices - and they were still leaning on UHPLC - often with very fast gradients and the discrepancy was a little jarring. 

Time will tell, I guess! I just wanted to drop this overall impression while I had a second - not long enough to write anything more involved. 

How much ion mobility would be necessary to replace the quadrupole?

 

Conflict of interest statement on this one - I'm both a compensated member of the Proteomics Advisory Board for Mobilion Systems and the first author of this preprint met weekly with me for 4 or 5 years while getting her PhD. I'm even more impressed by her work than by the quality of her education. ;)

Okay, but here is the real question and I know I've certainly wondered about this..... at what point would ion mobility and quadrupole isolation overlap? Or maybe this? Could you crank up your ion mobility resolution to a point you could widen your quadrupole and get equivalent data? Or if you had 1,000 resolving power ion mobility, would the quad just slow things down? 

Those are the questions tackled here. 

The short answers appear to be - absolutely, yes, and maybe. There is certainly a point where you can get enough ion mobility to get the same isolation you would get with a quadrupole alone. On the SLIM system the coisolation can be limited to about the coisolation of a quad running 5 Th windows. If you want the cleanest data you've ever seen, SLIM + 25 Th windows would get you there. And....yeah... in the world of wide window DIA or wide window DDA and now wide window PRM...these numbers are relevant. 

New Exploris confirmed! And...print cartridge LC columns....?

 

Just stealing Chris's post! 

Bonus, if you see Ian watch what he's eating and model your diet after it. He's at least 114 and he's out there on stage giving talks at 8 in the morning? 

Oh! Website went up on the Astral 2 "Zoomie"

270 Hz and the resolution to do TMT 32-plex (I thought it was 35? Does anyone know?) Big question should be - can it do the 32-plex at the highest acquisition rate? Or do you need to slow it down? 

We're going to see a lot of 300SPD proteomics on it, I guess. Probably with a 4cm column so everyone get excited to do label free quantification at 2 scans/peak! Still better than an aptamer! 

Informed DIA allows transcription factor quan in single cells? Also, new Astral?

 

If you've seen me talk about single cell proteomics, I like to point out the proteins I can see (50,000 copies and up) and I make a joke suggesting you come back in a decade if you're interested in transcription factors.

They turn on transcription at your DNA...you only have a couple copies of each gene. In many, most cases, we only expect a few copies of each of those proteins in each cell - certainly not tens of thousands! 

Yes this preprint demonstrates measuring them in global single cell data! 

How? Well...it's complicated. But basically it looks like we've got MS1 - DIA - and then what appears to be heavy isotope triggered wide window PRMs all in one cycle. I *think* the instrument sees the heavy spiked peptides and then moved a PRM window over to where the light should be. 

For you hardware nerds you may be interested in the fact this preprint seems to feature an Orbitrap Eclipse, and Orbitrap Asstral and something called an Orbitrap Astral Zoom, which might suggest we'll see an Asstral 2 at ASMS tomorrow? I guess we'll see! 

The first author will be presenting it as poster WP724 (Wednesday? Poster 724? I think!) if you're in Baltimore! 

pre-ASMS media drops - two new TIMSTOFs confirmed!

 

Hardware is already leaking to the media? Here is the first one - and a surprisingly well-written article with insight from a bunch of early adopters and/or beta testers. 

Oh crap, and my news feed was like "oh, so you like mass spectrometers? First I've ever heard you mention that... here's another! 

Looks like it's a top-down focused instrument with MSn(?) capabilities and some electron transfer thingamabobs, but I'm excited to see what it can do in the area of PTMs in bottom up as well. There area lot of PTMs that are just no fun at all to do with MS2 alone. 

I wonder if they had to get Disney's permission to use the same name or if it's so far from the meaning that there couldn't be confusion and trademarks wouldn't apply? 

 

pre-ASMS drops coming in! Seer system can prep 1,000 plasma proteomics samples/week!

 

Read more here! 

It sure makes the Illumina protein prep 384 samples/week seem even sillier.....

Electron capture increases protein stability by 60x - but you'll never get to use it!

 

Heading into ASMS shortly and there will be lots and lots of places where business and science interact. We'll see a load of vendor talks where the two things are clearly working together to make things better... but we'll also see... the other thing.

Case in point - 

An e-Mission EXD cell equipped after the quad on an Orbitrap UHMR can stablize the hell out of large proteins going into the Orbitrap for "single ion" detection (it's not really single ion, it's crappy signal measurements for deconvolution and enough combined to make it deconvolutable and since the mass accuracy is phenomenal it works AND it's way way way more sensitive than any other FTMS based intact protein measurement method - it's brilliant but the method name confuses people a lot). 

You know what it needs? Proteins to be 60x more stable!!! (It's not as much stability as charge reduction but that's enough nerd talk for now!) Want one? 

Too bad! E-mission was acquired! 

Man, they would still sell those products, though, right? 

Increase your multiplexing by adding the time dimension (and more nanoLC columns!)

 

My first thought when I saw this new preprint was something like "I've been working on these IRB documents for my clinical trial for waaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaay too many hours, cause I can't figure this out at all." 

I honestly might still not get it, so the less I type the better, but this is probably the best figure for driving the idea home in the preprint. 

I do think it is an interesting idea, and I'm all about getting extra throughput. Multiple samples running with time delays are demonstrated as being deconvolute-able using retention time re-alignments. 

From a practical standpoint - 

Since 2020 I think roughly 30% of my nanoLC columns have arrived in a completely non-functional state. The ones above just weren't glued at the ends so the capillaries just fall right out at the fittings. 

There is a vendor that produces nanoLC columns that have been famously reliable and successful over those few years, but - in my hands- the columns are amazing for about 2 weeks before they overpressure.

The thought of relying on 3 separate nanoLC columns performing properly chronologically seems just about as likely as me winning the big Powerball jackpot thing by buying 3 tickets. I've never gambled in my life, but I'm assured the odds aren't great of that happening. 

Spatial proteomics across laters of skins investigates psoriasis!

 

Okay - so who else was limited in their thoughs of how you could apply spatial proteomics in thinking of slicing being in 2 dimensions? 

This application find altered cholesterol metabolism in psoriasis patients compared to healthy matched controls at different layers of skin with the biggest differences being observed as you go down (?) 

LCM used a Leica 7 and LCMS was EasyNLC 1200 on a TIMSTOF. DIA and DDA(PASEF) were used and at least some of the data was analyzed in FragPipe. 

Really interesting stuff across the board. 

Last minute deadline for HLXMS Travel Awards to ASMS 2025!

 

Applications are still open - UNTIL TOMORROW! 

Chronic alcohol consumption alters acetylation in mice!

 

This is an extremely thorough analysis of the alterations that occur in the metabolome and proteome following chronic ethanol consumption.

On top of the proteomics - which involves both protein turnover analysis and acetylation enrichment multiple assays focus on absolute quantification of different panels of metabolite. Super solid work.  

DIA-BERT - GPU enabled DIA analysis!

 

I can't test this yet, because....ummm.....this is a pretty impressive suggestion for what GPUs you should use.... I honestly didn't know that there were 40GB GPUs! 

But wouldn't it be amazing to do DIA data analysis on GPU?!? Try running DIA-NN by giving it 12 cores, then giving it 24 cores. These tools obviously scale well, right? My old 1080TI in my desktop at home is like 10 years old and has 3,500 cores!

Let's run DIA on GPU!! And this one has a graphic user interface! 

MAP-MS! Get more Orbitrap dynamic range for free!

 

I had to sleep on this one and then find out that the coolest new toy I received for my lab needs to charge before I can try it out --

(not this paper - the fact I have to charge a device that I think will dramatically improve life in my lab and that while charging it I realized it has instructions I'm supposed to read) And now I think I get it. 

Remember BoxCar? I vaguely do, but it worked in sort of a similar way. Both try to take advantage of the fact that ion injection times on MS1 scans are really fast but Orbitraps are 

BoxCar works by chopping up your MS1 into lots of little DIA windows and then alternating between MS1 scans. 

Could you do something very similar but with great big boxes that wouldn't slow anything down at all? And if they were static could it be done with absolutely no obvious consequences? 

That's what MAP-MS appears to be! They study the distribution of trypitic peptides in humans to make their big window cuts and then multiplex their MS1 acquisitions to boost the lower abundance ions, reduce the highest abunance and everything seems to just work. 

On DIA experiments they get 11% more coverage without changing anything else. A lot of people will trade in a proteomics instrument for 11% more IDs, so that's pretty appealing. I do have concerns that maybe not every piece of sofware will love the data, but we won't know until we try it! 

DESI profiling of 165 proteoforms across over 10,000 single rat brain cells!

 

Wow. Okay, so I was a little (lot) less excited about the desciption of this technology when sea slug neurons were being profiled when the technology was first described here. Those sea slug neurons can be really convenient to work with since they can grow out on a plate to be easily 100x larger than typical cells. However - these are rat brain cells and these are going to be biologically relevant to more than just sea slug biology! 

They start with rapidly murdering the rats and getting to their brains and dissocation of the cells with papain. The cell suspension is then allowed to sink down and adhere to plates (fuzzy on this procedure, but it has been detailed in previous studies). Then those cells are fixed(?) in glycerol and some in ethanol (?) on the slides and they're ready for DESI analysis. 

DESI is like MALDI where you're moving spatially across a slide but the ionization is very different - from a mass spectrometrist standpoint the most important part is that you're multiply charging things. Here they can get the charges up enough that proteins are picking up enough charges to be detected in an Orbitrap (Exploris) running single ion methods.

It's a brief read, and a really interesting one. There is a really cool supplementary video and if you want to find what proteoforms were actually detected you'll want Supplemental Data 5.4. All the files are up on MASSIVE, but I suspect given the unique nature of the data it might be tough to make sense of them with the tools that I have.

Intact protein analysis of 10,000 freaking rat brain cells?!? 

More comparative isolation methods for Extracellular Vesicles for proteomics!

 

This is totally worth taking a look at - in every case this group starts with 100uL of plasma - and in every case the end with Suspension Trapping (S-Trap).

Do they cite the paper that gave the S-Trap method to the world? No, and that thing would be at about a zillion citations if everyone did. Otherwise, it's a nice comparative analysis. 

PFly - Is this the missing link in LCMS proteomics deep learning models?

Okay - so this one has bugged me (and a lot of other people for a long time) - we can do a pretty great job now of predicting peptide fragmentation (unless the vast majority of PTMs are involved). Supposedly we can do a solid job of predicting peptide elution patterns (exclusively from C-18 reversed phase chromatography). 

What has been missing is predicting what peptides from each protein will actually ionize (or fly). 

This has been tried before, btw -

- however, is often the case in academic software development, many of these lead to 404 errors or software that only runs in Windows 95 - or....well....they aren't very good. 

I'm a little sad to say this but when I did my normal round of sending a paper that I just found yesterday and was reading at lunch the responses were univerally ...skeptical at best.... but maybe this is finally it! 

Introducing pFLY! (I read it at lunch yesterday and it's faded in my mind a little but I'm just about 99.0% that the p stands for Pug) 

??