Astral vs TIMSTOF Ultra on real single cells? Just about evenly unreal numbers!

Maybe one of the coolest things about the EvoSep is how it really can allow us to minimize variables from one system to the next. You generally have the same column (or, at most, maybe 3 columns, which do make a serious difference, more on that some day), but the same flow rates, etc.,

Which can allow some head to head instrument comparisons that are hard to get otherwise! 

Without further ado - ultra low level samples - 

I don't have either, but I do have ProteomeXchange and 

and 

People who actually make their results publicly available! 

As another front end bit of usefulness both studies used this amazing single cell sample prep front end

Which generates AMAZING SINGLE CELL PROTEOMICS DATA. No question. Amazing. Is it the most expensive way you can prep a single cell today? Also yes. But before we update a preprint with a new clear winner for absolutely most expensive single cell proteomics study ever performed (you can probably guess what mass spec was used...) we'll pull down a crapload of files from these studies and process them with the same workflow.

Thank you DIA-NN 1.9.1 (also, after talking to Vadim I realized I should make separate library free libraries for each study, it does take into account whether you're using .d or .RAW when it makes the library. However, every other setting was left the same. Predicted off the same UniProt human library with all the same settings - and allowing DIA-NN to work out the appropriate windows for mass accuracy, etc., trying to be unbiased. Select file, select correct spectral library. Run. Wait. Again -boom - standardized data processing? 

As an aside - wow - Ultra files take a lot longer to run in DIA-NN. The mass accuracy isn't quite as good which takes longer and then it's got to do the IMS comparisons. I thought the Astral files were straight up crashing because they were done so much faster without that 4th dimension to think about at all. 

40SPD Whisper (nonZoom) for both. 

IonOpticks Aurora 15cm x 75um (which is Ultimate? Worth noting that they've updated their naming protocols for columns recently.) 

You're talking absolutely neck and neck here. Edge goes to the Astral by a tiny amount maybe? I'm getting 3,200 protein groups on the Ultra(1) and about 3,400 on the Astral per cell. HOWEVER, the paper using the Astral is using HeLa cells which has a higher total protein content than the cells the Broad used in this powerful demonstration of the budgetary power these two groups have. I think I processed 6 random cells from the Bruker and maybe 12 from the Astral. Largely because of the time constraints mentioned above. 

Both groups go to 80SPD in the study. Edge appears to go to the Astral at 1,200 protein groups/cell and 850/cell in the Ultra. Again, that might be the larger cell, but it is very difficult to tell. HeLa has this really useful smooth protein distribution, particularly in the high end, compared to most other cells which is why it's such a good cell type for demonstrating a proteomics method. 

Wait. No plots and error bars? Yo, this is a blog post I'm writing while waiting for espresso to do magic stuff to my brain. If I felt smart enough to fire up GraphPad, I'll start doing actual work. I actually did a decent job on this comparison at the time because I thought I was going to buy one of these big heavy things this year. 

Ultimately, these are always flawed comparisons so this is where I'm going to stop. You can go to ProteomeXchange and pull these down and process them yourselves. Both groups get much higher numbers than I do because they both generated actual libraries from their own data. We know that helps A LOT on the TOFs and generally less so with the Orbitrap data because we know what these predictors are based on - but the Astral isn't an Orbitrap and I don't thing we've got a good comparison? Today on what that difference is. 

At the end, though, it's really cool to see that we - as consumer scientists  - we have viable options for instrumentation. We can pick hardware because of how comfortable we are with the software interfaces, or for price or space considerations and then forget about the silly hardware arm's race and start doing biology with these things. 

ALSO - HOW CRAZY IS THIS??? The QC we had at the core I worked at before I went back to fail in academia was 200 NANOGRAMS OF HeLa and 2 HOUR GRADIENTS. I wasn't getting 3,500 proteins on my hardware with that! The fact you can process an actual sample that is 1,000 times lower in concentration and get 3,500 proteins in less than 1 hours is absurdly amazing. I think about this all the time. Where else can you say, yeah, we got 1,000 times more sensitive and faster in like 5 years? Absurd. And all signs seem to indicate we aren't out of this exponential increase in hardware capabilities quite yet.