Tuesday, July 2, 2019
Orbiter -- Double your TMT efficiency!
I just reread the short preprint describing Orbiter and feel that my rush job of going over it didn't do it justice at all.
In Synchronous Precursor Selection (SPS) MS/MS/MS (MS3) reporter ion quantification, there are 3 main steps (I'm trying to write out all my acronyms after some reader comments. There are new people coming into this field all the time! Don't kill them with acronyms!).
The first is the MS1 scan. All the masses that the Orbitrap can see.
The second step is/are the MS/MS (MS2) scan(s). Part way through the MS1 scan the computers onboard the Tribrid (Orbitrap Fusion 1/2 or 3) have enough information about the ions present that they start selecting the most interesting ions for MS2. The ion trap is crazy fast and in the remaining time left in the MS1 scan you can select dozens of ions for MS2. The combination of the high resolution accurate mass (HRAM) of the MS1 with the ion trap fragments is enough, in many cases, to fully sequence the peptide of interest. In SPS MS3 there is an additional step.
Ions in the MS2 are selected by SPS, isolated, combined and the subjected to MS3. High energy fragmentation is used to fully liberate all of the reporter ions from all of the ions SPS selects. The Orbitrap is used with a high (or what is now funny to consider "relatively high" because it wasn't realistically attainable by anything not all that long ago -- but maybe I'm getting old) resolution of, ideally 50,000 at m/z of 200 or so. This is way slower than the ion trap.
What Orbiter is doing is looking at the MS2 scans as they're being acquired, doing a search on them and providing information back to the Tribrid on what ions it should do MS3 on.
Yes. As I mentioned, the ion trap is crazy fast. So is Orbiter. Orbiter is using the Comet search engine. If you aren't familiar, it was written by a guy who had kind of a lot to do with something called Sequest. They share some similarities in their base functionalities, except he stopped working on Sequest a long time ago and he's still involved in improvements on Comet. (Other people have made improvements on Sequest, but I've never heard of a comparison between the two that Comet didn't win). The focus here is to do the search accurately AND FAST. Orbiter can complete a yeast search with some pretty loose tolerances (less restrictive tolerances typically increase search time) in 5ms and using all human UniProt + Isoforms in 17ms. This captures it better --
And it isn't a sloppy search, either. It's modeling peptide quality using a new and faster mechanism called a multi-feature linear discrimination analysis (LDA) that takes into account the XCorr, mass accuracy, percentage of matching fragments (this is a not a feature in classic peptide search engines, btw, but increasingly common, thankfully) and other features to tell good peptides from bad peptides. The Comet E-value is too slow, but the LDA method seems to be a huge win all around for speed and accuracy of the search.
The limiting factor here is really the relatively slow MS3 scan, which at 50,000 resolution takes 86ms. As long as Orbiter can figure out the peptide and what MS2 fragments from that ion that should be selected and the Fusion can select them and get them ready before the 86ms of the previous scan has elapsed, then this is a complete and total win.
What's important to note here is that you do not get more MS1, MS2, OR MS3 scans with Orbiter. What you get is far less MS3 scans that are wasted on junk that just happened to slip through without the intelligent filters of the Orbiter.
I don't have any Orbiter RAW files yet, but based on files I've got from ProteomeXchange from Lumos MS3 the normal way (there are Lumos files in the PRIDE repository as well as Fusion 1), such as this study, and comparing them to the files from this QE HF-X TMT MS2 study, MS2 based quan on a benchtop is still faster. You get more MS2 scans than the Lumos gets MS3. Even with the improvements on the Fusion 3 Vader system, I don't see where it could make up for this speed gap.
However, with no real time searching, the HF-X is fragmenting a lot of things that just slipped through. Monoisotopic precursor selection (MIPS) which is probably called "Peptide Match" on your system helps a lot, but it still allows stuff to get through that you aren't going to be able to sequence. So the number of matched peptides per unit time is going to be way higher with Orbiter.
Comparing SPS MS3 with and without Orbiter -- Orbiter allows practically the same number of identifications in HALF the time.
There is still just one downside in the back of my head that I'm concerned about. And that is individual genetic variation and PTMs we don't know about yet. Comet/Orbiter is fast enough to take in a surprising number of modifications and keep up performance (can NOT wait for hands on. I have some ideas I want to try). But what if your target organisms have a cool mutation that isn't found in your database that you don't know about yet. Or, simpler to think about, what if the thing that is really cool about your organism is a PTM that you don't know about yet? What if it's a sulfonylation, for example? If Orbiter doesn't know you're looking for a sulfonylation, is it going to skip all the peptides that have that modification?
I don't mean to end on a low note for what is an incredibly brilliant bit of coding that leads to somewhat unbelievable levels of performance improvements, but it is something to think about.