Only one day of work before ASMS Atlanta!!

If BiorXIV is any indication -- this year is going to be legendary. I...should...probably...pack..

For anyone who doesn't follow John Yates III on Twitter, he's been posting all sorts of protips and trivia for Atlanta. You can check them out here.

What was I doing again?  Oh. Finishing a poster and finding socks for ASMS ATLANTA?!?!?

Q Exactive Tune 2.11 has a cool button for the HF-X!

Did the HF-X have this button before? I jumped from 2.8(?)9(?) on this box to 2.11.

What's it do, other than use my marketing team's absolute favorite color that isn't a purple?

It only allows resolution where TMT10 or 11-plex will work for MS2 (30k, 45k, 60k).

It automatically sets your first mass so you can't forget and miss that pesky tag you paid for!

It uses 0.7Da isolation on the MS/MS? Sure! Works for me! Let's eliminate some of that coisolation background stuff.

I popped over to the HF to see if it was the same? Nope. Boo. You do have the 45,000 resolution option that was enabled in 2.9(?) but you have to set the settings yourself, or download the methods from www.LCMSmethods.org which has QE HF TMT methods good to go (2.7/2.8 right now)

Finally some hands on with Fusion (Lumos) Tune 3.1!

I'm not 100% sure when Fusion 3.1 came out (maybe last fall? I've lost all track of time), but I finally got some hands-on with it yesterday! 

Quick impressions: 

1) EASIEST Fusion upgrade path yet! That installer is great. We only had a problem because, for some reason, two different versions of Xcalibur had been installed previously on the PC. The installer executable would try to uninstall Xcalibur, see two and just kind of freeze. If I was going to do this again, I'd probably go to Windows "Uninstall a Program" wizard and do that part myself, just to save myself a couple of reboots. 

2) FAIMS is super easy to set up in the methods and you don't even have to think about it, many methods already exist that have the proper FAIMS settings! Since I have acquired some sort of damage in my brain in the area where other people can understand ion mobility stuff, this is 100% critical! 

While I'm highlighting stuff -- the ACTIONS pulldown is a little thing, but I think it is a really nice clarification for when you're building a method with multiple experiments in a single method. 

I think starting in Fusion 3.0 you could directly import someone's method from their RAW file. This functionality carried over and is ultrahelpful if you're hanging out in a lab that has both generations of the instrument. 

If you import RAW files from Fusion 1 to Fusion 2 or vice versa, you get a popup with a warning that you may need to change some values manually. 

 As far as I can tell, the only thing we had to adjust was the RF% (the number is lower on the Fusion 2 due to it's improved ability to get ions in without cranking up RF)

3) The new method templates present are great! There is a premade one for single protein characterization/biotherapeutics that I particularly like. Now -- this isn't an issue for these instruments, they are 100% full time dedicated to proteomics, but I lost all the premade small molecule templates that were there in 3.0. I'm going to guess this was something that I did, or they need to be unhidden by some button somewhere, but I couldn't find them.

EDIT (less than 1 hour later) yeah...I'm dumb....right above the templates is the pulldown for "application mode". Shoutout to the incomparable Darryl Hardie and his ability to read! 

More DEEP LEARNING for Proteomics!!

I'mma just gonna leave this here.....

Here is the link to the Kuster/Wilhelm lab paper. I was so pumped about this one last year that I got to an ASMS talk that was at 9 or something equally absurdly ridiculously early in the AM. (Learning is based on ProteomeTools -- maybe even the stuff that the rest of the world hasn't got to see yet? Can't wait to find a hole in my schedule today to actually read it!)


And the second one? I didn't know it was coming, but other people on Twitter said they'd seen it at ASMS as well. Probably means it was also in the morning or on day 2 when all the phosphorylation sites in my little brain were already fully occupied.  This may have some far reaching implications for both DDA and DIA -- what if we could massively improve the prediction algorithms for spectral libraries for DIA? There is a tremendous gap in the data quality that you obtain when you match real DIA spectra against experimental and predicted spectral library data.

The first one is called PROSIT!  As always, I'd like to thank Google Images and all the bored people with meme generators who make the majority of content on this blog possible.

Wait. What? What the Michelle? Heck is this?  It appears to be very very related to this thread and needs investigation!

Ummm....IonBot is here....

SurfaceGenie -- Time to finally make sense of the cell surface?!?

This might be the new record holder for the most forwards I've made of a single paper before I finished reading it. I know so many people who need to check out this resource -- and I bet you do as well as soon as you see the power of what you are looking at.

You should check out this preprint here!

Want to just bumble around on an easy web interface first and let what you're looking blow you away first?  Go to CellSurfer.net and pop over to SurfaceGenie.

Whew! I was 600+ words into a ramble about why I'm excited about this, realized I wasn't doing this justice at all and deleted it!

Here is the summary of why I'm pumped -- SurfaceGenie helps us find and understand the proteins on the surface of human cells. We know VERY little about what is going on there, but the little we've been able to put together so far has (basically) made these amazing new drugs possible.

Sure -- the inside of cells is cool and all...but the outside?!?!? if you know what is going on there you can target the destruction of the ones you don't want hanging around. This is both a big step forward for human proteomics discovery and for that "precision medicine" thing we keep hearing about.

Measuring the accuracy of SPS MS3 in Proteome Discoverer 2.3!

Proteome Discoverer 2.3 has been out for a couple of months now and I've been too busy to even talk about it(?!?) which is more than a little crazy, probably. Loads of greats new features -- but let's talk about my very favorite one -- directly measuring how well Synchronous Precursor Selection (SPS) worked in your files where you used it!

After processing you've got your Peptide Spectrum(al) Match(es) so you know what (typically) b and y fragment ions truly came from your peptide of interest -- and, in your scan header, you've got the MS2 fragment windows that SPS used to acquire your ions for MS3 based quantification --

PD 2.3 integrates all this and can give you a metric of the percentage of the SPS fragments that the Fusion or Lumos decided to use that actually match the fragments that you used for your PSM! You can even set filters in your consensus where you throw out the quantification of any ions where the SPS didn't select the percentage of signal from ions that you decide is a good cutoff! REALLY concerned about that quantification accuracy? Then only keep ions where at least 90% of your quan signal came from fragments that were really your peptide!

We thought this feature was cool enough to pull a bunch of ProtoemeXchange files down and evaluate SPS accuracy. If you're bored, there is a bioRXIV preprint with more details here (and an unfortunate figure numbering glitch that is 100% my fault (I'm dumb, sleepy, and a bad writer)-- but this is what we found -- and it's mostly normal -- with a big surprise that's made me feel more deranged than usual the last few weeks:

1) Simple samples like yeast or prefractionated samples like offline fractionated and concatenated human lysates (as long as you do a good job with the fractionation) has high levels of SPS selection accuracy. Files like this also have low levels of MS1 --> MS2 coisolation interference, so that makes sense.
2) SPS works just as well if you're doing your MS3 based quan in the ion trap or Orbitrap (MS3 -TMT6 in the ion trap still seems like a better idea than MS3-TMT6 in the Orbitrap
3) Single shot human stuff --  does not work nearly as well. Too many times your MS2 window is a jumble of fragments from 2 or 3 different peptides originally selected from the MS1 together and SPS can't tell what your target is.

All normal stuff, right?

4) Suprise? Okay -- so -- honestly, I might just be stupid. For real. However, if you look at the MS2 coisolation interference that I consider a critical parameter when I'm doing MS2 based TMT quan and the SPS Accuracy, they don't directly correlate. Just because an ion has a friend that is right in it's retention time and m/z window and is using up it's MS/MS fragmentation window space -- like this one --

(w(inner) is the ion I want to quantify -- l(oser) is the one I don't care about. If you select any window around the w ion, you're going to get stuck with the a bunch of l fragments, right? A lot of people doing MS2 based TMT would exclude that peptide from quantification -- can't trust that value, right?

You'd think that would be an MS/MS event where SPS would do a bad job -- and -- as best as we can tell from the 5 or so sets we tested (3 are in the preprint) there is no direct correlation between the 2 variables.  Why? No idea.

What's important, though, is that we've got a new way to make sure the TMT/iTRAQ signal we're reporting for each peptide is as accurate as possible!!

If you're doing at least 12 fractions of human stuff with concatenation and solid preMS chromatography, using an SPS filter isn't going to change very much. If you're doing single shot ultra complex stuff, using an SPS filter is going to cost you a lot of quantified peptides -- but this is the first way I've EVER seen to pick through that data and only keep the good stuff! Go PD team, go!

Now that we know how to do this, it should be pretty easy (unless the idea is trademarked and stuff, definitely check that, in case) for other programs to integrate it. The values are right there in the scan header!

The landscape of cancer cell line metabolism

Ummm....wow....

In big data studies you can generally go one of two ways, right? Loads of samples at lower coverage or fewer samples with extensive coverage. This huge and ambitious study did the former.

1) I massively respect the amount of work that went into this. This is almost 1,000 different cancer cell lines metabolically profiled for around 225 total metabolites. The controls and different chromatography methods used are so impressive that it doesn't matter that it's QTrap data. I'd normally --

-- if asked to validate an MS1 identified lipid that was qualified by a 0.3Da window -- but -- the power here is in the great n -- 1,000 cell lines??! I don't think Jeff Bezos could afford to send that many samples to the great big company that does all the awesome metabolomics, so this group did it in house with their 2 QTraps.

MASSIVE BONUS for any of y'all thinking about doing metabolomics. The supplemental material has all the settings and transitions for the 150 or so polar metabolites (these are surprisingly hard to find out there in the world) only the lipids were knocked out by what appears to be MS1.

MORE MASSIVER SUPER BONUS -- All this data on these 1,000 cell lines is available online at the Broad (which -- I'll remind you -- is pronounced like "TOAD") -- you'll need to register and make some promises about how you'll use the data and that you'll report the errors back to them -- but WHAT A RESOURCE!!  The metabolomics is just the top of the iceberg. There is genetics stuff galore!

BioRender! EASY and Impressive scientific figures!

Shoutout to the BioRender team for providing me a high res version of this cool video above so I can use it here!

You can check out Biorender here.

First off, there is a commercial version of this program. Please see my Disclaimers statement over there --> somewhere.  Importantly, there IS a free version and that's what I'm looking at!

I suppose there is some section of the human brain that is supposed to appreciate art. Mine has never worked. I've groaned my way with friends through Art museums, including some famous ones in Europe and wished I was doing something cooler, like walking barefoot on syringes on the streets back home in beautiful Baltimore. I think this reflects on the few papers I've done in my life. I'm going to throw something together in Excel or Powerpoint, ignore friends who will say things like "...you're so funny, you'd never really submit an Excel graph, right?!?!!"  and send it out the door.

This piece of software look like a way to get something much much nicer and somehow EASIER than Powerpoint!

Virtual lock masses!

Time Of Flight (TOF) instruments can be ridiculously fast. Compared to today's FTMS (mostly Orbitrap) instruments, they compare poorly in two pretty serious regards.

Is this new paper in biorXIV something that could finally lift one of these limitations?

My blogger-at-5:30AM interpretation is:

1) Sensitivity -- with the exception of TIMS, there is no way to accumulate signal before shooting the ions down a flight tube -- this is a generalization, but I expect around 100x less signal on a TOF compared to even the oldest Orbitrap systems (also keep in mind that more TOF resolution = less TOF signal since the ions are traveling farther in an imperfect vacuum system.)  This is even worse, considering that I sometimes acquire signal in C-traps prior to Orbitrap analysis for 1000ms (or more) when I want to prove that a QE can exceed even quadrupole LOD/LOQ

2) Accuracy -- again, more generalizations, but most TOFs without consistent lockmass are accurate to one decimal place, where Orbitraps are at least to the second decimal and almost always to the third. (TOFs are also massively affected by temperature in the room, with some needing calibrated many many times/day)


Aha! Okay -- so I haven't ran a TOF in a long-ish time now -- but -- I know consistent lockmass sprays are in use that improve #2, but could you mathematically drop in a virtual value and get improved accuracy all the time?  These authors sure think so.

I tell you what. I was seriously proud of my brain as far as page 5. For real. I probably didn't actually know what was going on, but some combination of espresso and birds chirping at 5:30 made me feel like I did. Go brain go! I had some random thought about "Vince Carter's coming back for one more year, maybe there is still hope for you down this senescence water slide, old brain!" ...then...

...maybe it's time to walk the dogs and look at flowers....

Importantly -- yes, this totally looks like it works, though my dummies interpretation is that you need a lot of spectra to get the maths to work out, but once you have that pile of spectra you can use learning machines to massively improve the quality of the TOFs data.

Now -- I know people often think that the mass accuracy off the Orbitraps is probably more than we need -- and just about no one uses MS/MS lockmass on their Tribrid systems, even though that capability is there -- but -- is it enough? Will future bioinformaGicians look back on our data and wonder "why on earth didn't they lockmass their MS/MS spectra -- I need it for this (insert undiscovered chemical modification here)!??!" I don't know, but it's nice to think that we could improve the data we already have!

IPSA -- A handy tool I bet you're going to use all the time!

Sometimes you just need to step back, look at where your pain points are in your field and just put someone smart on fixing them forever.

Pain point? Annotating all those stupid fragments that your GUI output software doesn't feel like doing for whatever reason....or does but it looks awful....

Solution?  The Interactive Peptide Spectral Annotator! You can read about it ahead of print here. 

I highly recommend you read the paper. I will as soon as I have time to breath. Writing this and generating the meme took up all of my free time this morning. (Time well spent, in my ever humble opinion).

However -- if you are also pressed for time -- this tool is so amazingly intuitive that you can get going with it right now!

The online version is here!

The Github so you can use it and integrate it with your bioinformagical powers is linked in the paper.

I just punched in some random peptides into the online thingy here....

(..I mean..I was already on the site...but -wow, the output is super sharp)

MARMoSET -- Get publication ready data out of your RAW files!

OMG. I'm either sleepy and/or this is hilarious. Full song and video at YouTube here.

What I'm talking about, though is this great new paper in Press at MCP.

How much time have you spent trying to get data out of your RAW files and to a resolution that you could actually submit?

Bonus -- As someone who has spent a decent amount of time trying to come up with cool abbreviations for methods -- this is an awesome example.

P.S. -- Have we had the UrbanDictionary talk? If not, we should. Please -- before you publish that cool word you came up with for your method -- please -- check urban dictionary for it first. This is to ensure your new software or technique doesn't lead everyone looking for it online first comes across something terrible you weren't aware human beings were physiologically capable of doing. Definitely may have happened to someone in our field recently. Don't be a statistic, please. 

This is how Marmoset works -- and how they got the name!!

Focus on the spectra that matter, duh!!

How did I miss this!??!  Looks like I retweeted it months ago, but didn't read it?!? 

How much time are you spending processing PSMs from proteins like this?

12 minuted CE-MS/MS run -- 145 PSMs from Tubulin B4. Who cares about Tubulin B4? Literally no one who has ever lived on the planet earth. Okay -- maybe 6 people who ever lived on this planet.  I don't care about tubulin. Especially not enough to waste 145 MS/MS events in a 12 min CE-MS/MS run.

Why don't I care?
Because it isn't changing. I only care about quantitatively changing proteins. Or PTMs that are quantitatively changing. If the ratio is 1:1, no one cares about your dumb protein.

How do we do proteomics, though?

We identify everything -- then we quantify it. If we're really lucky our software does both at the same time.

What we should do is use something like Quandenser/Triqler (a DINOSAUR IS INVOLVED, which is another program, but that's okay) to find out what is changing first -- THEN ID IT! 

You can get this program from THE Matthew here.  This program works for label free stuff.

Unrelated project, but related idea --

If you want to do something similar for TMT/iTRAQ, you can get RIDAR from Conor Jenkins GitHub here. 

About time!!

Press release is here.

Great review of post translational modifications associated with aging!

Okay -- honestly -- I was absolutely following this review from a couple of years ago until it gets to the E.coli stuff. Which is probably awesome/relevant in some way besides making very pretty pictures, but the first part of it is definitely great.

Reviews of PTMs and aging and the literature that has found links between these? Fantastic! You can check it out here.

Effects of APOE on Brain Proteomic Networks!

Somewhere over there -->
I put together a guide a few years ago about what to do after you have proteomics data. I'm glad to hear some people have found it useful, but I know it isn't very good. Truth is...I don't really know what you do next to work out a mechanism if GiaPronto or Ingenuity doesn't say "It's this one!!"

This amazing paper in a journal I've never heard of brings new set of tools on stage and goes through them in painstaking detail.

I strongly recommend reading it in the HTML format over the PDF because having the references side-by-side makes it much easier to pop a tab open and figure out what they're doing (in case it's a word of stack of letters in a row you've never heard of).

...which...makes me wonder why everything isn't formatted this way!!

Don't get me wrong, the proteomics here is stellar. (They run a Fusion in HighSpeed mode using 50cm columns and 150min gradients and combine this with data from another cohort that used a QE Plus). Most of the methods and basic data analysis are based directly on the spatially resolved mouse brain study Max Planck did a few years ago, but its the amazing level of detail in both the experimental design and downstream analysis that makes this something I'm very glad I read this weekend!

Baltimore/Washington Mass Spec Area -- Let's talk about new software!

Ummm....so...thank you @the_ion_doctor for reminding me of this thing I agreed to do a year ago that I definitely wouldn't have forgotten about otherwise, probably!

In the area? Want to talk about data processing? Find our little group here!

Let's talk about --
OpenSearch Strategies (Fragger.TagGraph) and hybrids like MetaMorpheus
Second searching!
Using MS1 libraries
Scaling data processing way up with GPU and Cloud-based processing
And how to take all this lousy nextgen sequencing crap and make it into something that you can actually search with?

That's what the overview on slide 3 of my totally completed and well-orchestrated slide deck says we'll do and that probably won't umm..change...too much... Slides so good I'll post them, here after the talk.  Be careful when removing the limiters, though....

Register for the ASMS Skyline User Group Meeting!

There is no shortage of things to do the weekend before ASMS! However, if you haven't already signed up for awesome workshops the whole weekend -- registration is still open for Skyline!

You can register here!

1) Its in the Georgia Aquarium
2) If you're really super extraordinary, you might end up with something like this:

3) The speaker lineup is (again this year) loaded!

4) I might finally get to meet my collaborator -- schizophrenia proteomics expert Matt MacDonald in person, since he's doing one of the talks.

Related -- a study we both worked on just went live this week at AJP (classic medical style journal, I'm pretty sure it went into review in 2011....his group at Pitt did all the LC-MS work, so it's really good.)

PROTEOFORMER 2.0! Don't compete with RiboSeq, Assemble with it!

Gotta move fast on this one, but -- holy Unicron -- this is exactly what I was looking for. I've got proteomics and RNASeq and some PacmanBio stuff of some samples and this very frustrated researcher was explaining to me that he had "RiboSeq" of the same thing and I just kept staring off into the distance because I do that when I'm trying to assemble information I don't understand and it makes people think that someone is sneaking up behind them. And -- check this out!

Don't know what Ribo-Seq is? Me either! Here is a WikiPedia article.  My understanding is that it bridges the gap -- only things that are actively getting to the ribosome for translation are sequenced. It's the closest you get with probes and genetics before getting to the proteome!

Obviously this isn't new -- there appears to be a 1.0 version, for example -- but it's new to me on a totally new concept -- and it's so powerful that this group uses it to identify NEW PROTEOFORMS from shotgun data!

Bonus -- 1980s chemistry courtesy of poorly paid animators.

ANN-Solo returns with a GUI and GPU processing!!

Okay -- there are still some downsides to ANN-Solo in my mind, but they're mostly because I'm old and dumb. Here is an older post I made on this great software.

Downsides -- Python. The world's most approachable powerful coding software. You need to use it. If you've got neuroplasticity left you can probably learn it in an afternoon. Mine's all used up.  Fortunately I have Python experts around my lab! Two all the time and 3 on Thursdays!

Downsides -- Linux. Wait. You can use OS-X? Is that what Macintosh uses? I forget.

What am I rambling about? This new preprint featuring the return of ANN-Solo with even more power (and a GUI) 

Want a reason to get a Linus NumPy things all set up on something in your office?!?!

5.6 milliseconds to search an MS/MS spectrum against a spectral library! To get this fast, there is a catch, though, you have to use a Graphics Processing Unit (GPU).

GPU data processing isn't new. My old Waters Q-TOF used it 7 years ago. You can buy at least one commercial data package that uses it, and I think Darryl Pappin was messing around with them quite a while ago.

GPUs have TONS of cores. My 1080TI in the PC I'm typing this on has 3584 cores (they're called CUDA cores). Compare this to my CPU that has 20 cores or threads or so. However, each CUDA core in a GPU is weak and dumb and only capable of doing small tasks, like controlling a few pixels in a video game or performing the same dumb math problem over and over again in order to try and construct a block to win you a BitCoin. They have another big downside, as well, in that if you exceed what they can do, they aren't smart enough to stop the code. They "overflow" or something and the little core outputs gibberish.

ANN-Solo breaks high resolution data files into tiny little parts so that each little core only has to do a little -- then you can use the thousands of available cores to tear through files.  Compared to a decent CPU, they drop the time to process IPRG2012 data from 50 minutes to 6 minutes.

In the defense of the CPU, they use a VERY good GPU. The study uses the new 2080 GPU. That's a >$800 and pretty tough to justify if you're just using it for gaming....

Hey Ben -- this doesn't sound all that exciting....what's the big deal!??!

Have you seen some press releases at your university or facility talking about the "17 zillion (insert fake sounding number here) cores at the new High Performance Computing Thingy" that anyone an have access to? I bet you have.

Have you tried to use it? Were you surprised when someone complained that you used 48 cores for 24 hours and wondered what's up?  They're probably talking about CUDA cores and they probably don't actually have all that many of the cores that your software uses.

I bet the details are in the paper. I dunno, but in general you can link GPU after GPU together. What if ANN-Solo is the way to use all that HPC stuff? How cool would that be? And that's why I'm excited...

..groans...

HDX vs FPOP -- They're both better than NMR, but which one is the bestest?

Structural mass spec is advancing like crazy. Crosslinking keeps getting better -- with new reagents, better separation and improved MS/MS techniques and -- of course better software (1, 2, 3 )leading the way. (Here is an awesome recent review on all things proteomics structural.)

Crosslinking gives us a lot of power by letting us know what amino acid residues are close or reeaally close together (is this one commercially available yet?)

There are weaknesses here, though. Things can be close together without being within the space necessary for us to chemically lock them together. Conformations of relatively "low stoichiometry" (hey -- if phosphoproteomics can use this terminology as if it's okay for 15 years, let's keep bending it, it's almost a meme now!) are going to be impossible to see AND we learn nothing about modified residues or the outside of the protein structures.

Two techniques are improving all the time that can give you a lot more of this information -- and they go head-to-head in this study I left myself a note to read in September and just found.

Chances are you know more about these techniques than I do -- but I'm learning 'cause I think they're only going to become more important all the time!  HDX requires some arduous sample prep up front or the purchase of an add on system for your mass spec that does all the work for you. Deuterium can't get to the inside of your protein as effectively, so anything that gets labeled is on the outside. Boom! I know what the inside and outside of this protein and protein-protein interaction is like and tons of smart software exists that helps interpret the data. Workbench is a good example!

FPOP uses hydrogen peroxide and

...lasers to modify the outside of proteins, protein complexes, and -- holy shit -- have you seen this?!? -- even works inside whole cells!!  Study 1, study 2. The downside is that the modifications on the outside of the protein may be unpredictable. Better data processing, better resolution and accuracy data have helped make this easier, but it's still tough at this point.

Both of these techniques are better than NMR, obviously, because I don't have an NMR and the whole concept seems really old fashioned, and helium is not getting cheaper -- if you are a new lab starting out you may find that it's hard to get Helium at all (we had to go through 3 vendors and lie and say we're doing medical research (!!kidding about the last part!!) -- to even get tanks!)

 In the study that I originally started talking about (comparing HDX of the same protein to FPOP) HDX comes out on top. The comparison might not be the most fair, though.

1) HDX analysis was performed on a QTOF system
2) FPOP analysis was performed on a QE Plus system

I know someone who has an HDX QE Plus (Hi!) and I wonder if she'd get the same results with the same material? I think the higher resolution and higher sensitivity of system #2 is kind of seriously essential here, maybe particularly because the QE Plus for the FPOP used 300ms(!!!) of fill time for MS/MS, indicating that even with a pure protein collecting a lot of ions is critical component, which the TOF utilized in this study can't do (only one TOF kinda sorta can, right?!? I'm only putting question marks here because I've discovered there was some really interesting hybrid TOF technologies that were developed in the past that we all seem to have pretty much forgotten about).

So...it looks like FPOP wins! I like this result because HDX sample prep seems too finicky for me to ever get it right!