Proteome "phasing" revealed by transcriptomics!

Okay....want a reason for a whole lot of those unmatched spectra in your human LC-MS/MS runs?

Check out this paper in Nature Biotechnology from Hagen Tilgner et al.,!  In this study they used transcriptomics (in this case, long-read RNA-Seq) and looked at different tissues. Turns out there is splicing everywhere!

Splicing? Well, that's when DNA that should be over here making this protein -->
ends up hanging out
with DNA
that's way over here -->
(who says blogging can't be hi-tech!)

and you end up with a transcript (and therefore a protein!) that, from a purely DNA perspective, TOTALLY SHOULDN'T EXIST AT ALL! (Definitely isn't in Uniprot/Swissprot...!)

Sorry for shouting.  I'm excited. This paper focused on mouse brains and found a whole ton of these things. In regards to some of the recent discoveries in brain proteins, maybe this isn't that big of a deal, but tissue-wise, holy cow....should we be using a different FASTA file if we are profiling liver tissue than if we are doing tissue that came from brain samples?  Sure looks like it!  Heck, if nothing else, its another great argument for PROTEOGENOMICS!  (Yeah, I'm shouting again!)

Worth a read. Sorry it isn't open access. And sorry if this is jumbled. Its kind of late and I've been excited about one thing or another for the last couple of days.

Assessment of longitudinal interlab variability

This one is pretty cool .

What happens when 64 different labs submit BSA samples that they run every month for 9 months and people sit down and assess the data?  Sounds like an ABRF study to me!

As we've come to expect, intralab variability (same lab over the 9 months) was smaller than interlab variability (from one place to another...I get them mixed up). That makes sense.  My LC my mass spec, I'm going to keep it pretty consistent for 9 months, compared to the way I run it versus those wackos over at Whats-It-Called University.

Variability among all the samples really doesn't look all that bad. It his, however, a single protein digest -- so we'd kind of expect that. Sampling of 100,000 peptides from a normal mammalian line might be a more sensitive indicator, but I still think this is a promising measurement.  As a field, we're getting better all the time!

Interestingly, the real outliers seem to show up right after LC-MS preventative maintenance (PM). And this makes sense, too. If you've had your LC open and changed some thingies in it recently then peakwidths and retention times might have shifted a bit pre- and post- opening it.  Sure does emphasize the frequent use and recording of quality control standards, particularly after maintenance and things.

Oh yeah, and this paper is currently open access in early release format at MCP here.

Can you use the SMART digest kits for proteomics?

Okay, y'all know I've been trying to find a good, easy, and reproducible protein digestion method to get behind. And I've mentioned the SMART (previously Perfinity FLASH) digests kits before. The big question that is floating around is: it works for single proteins just fine, but how well does it work for proteomics?

According to this paper it works pretty darned well. Now, this is just one paper and all, but I really can't come up with a reason that a method that digests one protein wouldn't digest a whole bunch of 'em and do it well (so long as the protein to enzyme ratios aren't all wonky).

Sure, its N=1, but it sure looks like it works!

First impressions of the free LFQ node from OpenMS

Over the weekend I got to finally toy around with one of the cool free nodes from the Kohlbacher lab that we can install into Proteome Discoverer 2.0.  The LFQ is short for "Label Free Quan" and the nodes are freely available to download for anybody here.   Now, before I go forward I should probably reiterate something that is on that page. These are 2nd party nodes and these won't be supported by Thermo's Proteome Discoverer team. Questions should be directed to the node developers. Fortunately, they seem quite straight forward!

Here are some early impressions of the nodes.

1) They are easy to install. Download the file from SourceForge, make sure all Proteome Discoverer versions on the PC are closed and run the file.  When you reopen, the nodes are there!

2) The node developer's even have workflows ready for us! There is a Processing Workflow and Consensus Workflow. Which is great! Cause, honestly, I wouldn't have thought to set them up as above....

3) Interesting note.  SequestHT and Percolator are mandatory. Gotta have 'em or you won't go anywhere, it seems.

4) LFQProfiler appears to multithread.

Windows performance loggers are always kind of hard to interpret, but all 8 cores on this desktop appeared to be doing something when the LFQ Profiler kicked in.  In the consensus you can actually tell that LFQ node how many cores it is allowed to use!  On other runs, it looked like I was maybe only using 4 cores, but this really isn't a good measurement.

5) Disclaimer here: I've got like 10 versions of Proteome and Compound Discoverer on my desktop because I have been alpha/beta testing them for years. I've got some versions that are locked down for different projects so my working environment is probably sub-ideal. But... I'm gonna be honest here, and I'm likely doing something wrong, but I'm finding the node a little difficult to integrate into my workflows, in an odd way. I keep getting "Execution failed" in my Administration tabs, but the failed workflow can be opened and looks just fine. I do have to unhide my Intensity but the numbers are here and it looks like it ran real fast!

So...first impressions. The LFQ node installs easy, has convenient pre-made workflows (additional downloads required) and seems to run fast. More analysis required to see how it works, but its Sunday and this is all the PD I think I'll do today!

UPDATE: 12/2/15.  Downloaded the LFQ nodes and they are AWESOME!!!

More statistical analysis of PSM FDR from the Qu lab!

We have an awful lot of search engines these days and we have almost as many (more?) ways of working out our automatic false discovery rates.  The Qu lab seems to have stepped back and said, let's try to sort it out, meaning, which FDR is more appropriate for large datasets -- and when?

This is a heavy analysis of three different search engines available for running in or through Proteome Discoverer as well as an analysis of what false discovery rate algorithm/method or filter will leave you with the best possible results.  Interestingly, the answers appear to be very analyzer and fragmentation-type dependent.

I'll leave this here for you guys who took more maths!  The answer appears to be...there is no easy answer...these are things we're definitely need to spend more time working on as proteomics moves further and further into the BIG DATA world.

You can find the abstract for this (paywall) paper here.

Histone post-translational modifications in monocyte-derived macrophages

(Picture credit: Laxmi Iyer, original url [unrelated to this article] here.)

It turns out that most of the histone work that has been done out there has been done on mouse cell lines or immortalized human lines. While this is undoubtedly useful information, immortalized cell lines tend to be kind of messed up and we all know about the plus/minuses of studying mice.

The Ciborowski lab has a plan of long term, in-depth studying of histones and their post-translational modifications of normal human macrophages. In this first study (available here, open access) they work on establishing their normal, baseline conditions for resting macrophages. Once they get that, they can go on to further studies.

For this analysis they are primarily using an LTQ-Orbitrap XL with ETD and employing both CID and ETD fractionation. Surprisingly, the majority of the information being obtained for PTM matches is not coming from the ETD. This is likely due to the lower speed/efficiency of the earliest ETD system compared to the ones I normally get to mess around with. It does, however, contribute meaningfully to the study. This is a nice clear study but I mostly highlight it here because I'm very interested in what they are going to do next AND how this data is going to line up with other well established histone PTM datasets we have from other models.  So...this post is kind of to remind myself to check back on these guys later...sorry...

EuPA 2016 --- Turkey?

The theme of the conference is "Challenge Accepted. Standardization and Interpretation of Proteomics." Ummm.....YES!!!!  And the lineup of speakers is already AWESOME!  You can check it out here.

New free label free node for Proteome Discoverer 2.0!

2 hours in and complete brain overload here at HUPO!  So much good science out there in the field!

One important side note that I think you guys will like, though. The first free OpenMS nodes for Proteome Discoverer 2.0 are now available for download. The first is LFQ for label free quan!  The second is a workflow in OpenMS I'm unfamiliar with, but will read up on ASAP.

You can download these here.  (Let me know how you like them! I can't wait to give 'em a shot!)

Quantitative thermal proteome profiling!

About a year ago a paper came out that introduced me to the concept of Thermal Proteome Profiling.  While this concept will likely have several different applications, it is definitely really good at figuring out what proteins a drug is interacting!

This month, a brand new paper out of the Savitski lab takes this idea another step further. In this work they use TMT10plex reagents AND thermal proteome profiling to determine what proteins are being directly affected by certain drugs.

Being a Nature Protocols paper, the methodology is set out completely so that any of us can go right out and replicate it. This team also developed software for Python and R that can be used to process the data and they make this all available.

Are you stuck on what that stupid drug's mechanism of action is? You should probably check out this paper!

NeuCode labeling nematodes!

As depicted in the clip above, nematodes are hungry little guys. In this brand new paper in press at MCP from Rhoads and Prasad et al., we see a new way of taking advantage of this trait.

These Badgers fed nematodes NeuCode labeled E.coli and, voila!, NeuCode labeled nematodes!

Now, I know I've rambled on about NeuCode in this blog a bunch, but if you are unfamiliar there is a good description in this GenomeWeb article here.  In a nutshell, its very much like SILAC except by NEUtron endCODE(ing) the mass discrepancies between the various channels are very very small. You are limited by the number of channels you can use by the maximum resolution of your instrument. More resolution = more NeuCode channels.

(I stole the figure above from this open access paper here.)  P.S., the technology has been progressing significantly since the original study. I saw a slide a while back that suggested 40-plex is theoretically possible.

In this study, the nematodes are studied with an Orbitrap Elite that is running 480,000 resolution at the MS1 and 30k resolution at the MS/MS (which they refer to as "medium" resolution! man, I love this field!!!)

Now, you might think 480,000 resolution? That's so slow, they'll never identify anything that way!  What did they see? The top 50 most abundant proteins? Well, they did a little better than that. This might be the single most extensive proteome of the nemotode out there. Along the way they did phosphoproteomics and also worked out some of the key regulators or stress response in this important model organism.