iiinova — where your digital self keeps evolving

Christopher Rose @cmichaelrose.bsky.social · 03/06/2026

In ccRCC cell lines, MIRA-MS quantified 97–107% more non-canonical peptides and 45–89% more canonical peptides vs. standard acquisition. Great collaboration with @nesvilab.bsky.social

000

Christopher Rose @cmichaelrose.bsky.social · 03/06/2026

⚡ MSFragger-RTS — fast, fragment-indexed real-time search including nuORF databases 🤖 Salud — a Prosit-derived AI model for real-time fragmentation + retention time prediction 🔬 Real-time FDR rescoring UI to improve canonical vs. nuORF assignments

110

Christopher Rose @cmichaelrose.bsky.social · 03/06/2026

Non-canonical (nuORF) HLA peptides are promising cancer targets, but their low abundance makes them hard to find. MIRA-MS combines:

110

Christopher Rose @cmichaelrose.bsky.social · 03/06/2026

Excited to share our new preprint on MIRA-MS — a real-time AI-guided acquisition strategy for deeper immunopeptidomics!

171

Christopher Rose @cmichaelrose.bsky.social · 03/06/2026

Writing from the airport. Sad to miss it, but out preprint just dropped in time! Have a great discussion tonight! www.biorxiv.org/content/10.6...

biorxiv.org

041

Christopher Rose @cmichaelrose.bsky.social · 30/05/2026

I already landed in Sam Diego and had a great day at Legloland today. Check out these #ASMS2026 presentations Genentech and the Proteomics department from the recently formed Translational Genomics and Proteomics Center. Very proud of all members of the department for their showing this year!

060

Christopher Rose @cmichaelrose.bsky.social · 06/03/2026

Very cool - glad to see folks picking up from libraries like that. We will be sharing more of our libraries soon linked to an upcoming publication, hopefully they will be equally useful for folks. Our approach in inSeqAPI is a bit complicated, but very customizable.

130

Christopher Rose @cmichaelrose.bsky.social · 06/03/2026

I am curious, are you writing your own IAPI program? Or using one of the (few) options out there?

120

Christopher Rose @cmichaelrose.bsky.social · 25/11/2025

I am looking for a PostDoc to join my group at Genentech to drive the development of proteomics technologies to fuel target and therapeutic discovery and development. Check out the link below! careers.gene.com/us/en/job/20...

12324

Christopher Rose @cmichaelrose.bsky.social · 30/07/2025

I suppose that could be true. But having spent a large amount of time over the past 1.5+ years trying to understand DIA data and its analysis, the thought of adding in time multiplexing seems ambitious. We already see wild variations between softwares and versions with the “simple” data we have now.

160

Christopher Rose @cmichaelrose.bsky.social · 30/07/2025

What’s interesting to me is that more than one group is taking this approach. I’m still skeptical, you are spreading out the impact your high abundant peptides across the run, impacting low level peptides. Not to mention peptides unique to a sample. But interested to see where this approach goes!

160

Christopher Rose @cmichaelrose.bsky.social · 25/01/2024

As of now the only way to remove 126 is to take reporter ions to ion trap and isolate there. We did that in our Nature Methods paper, but you lose some 127. Other way would be to raise RF amplitude enough to create a low mass cutoff. Interesting idea, but would require some instrumentation changes.

020

Christopher Rose @cmichaelrose.bsky.social · 31/10/2023

Check out this Postdoctoral position! Two great labs at Genentech to be co-mentored by. You get to work in our department. You can see me as much (or as little) as you want!

012

Christopher Rose @cmichaelrose.bsky.social · 27/10/2023

I imagine the coverage penalty will be pretty severe. For single cell, Trypsin peaks alone will cause significant issues when you spread them out (which you will with a tag solution). Also forgot to mention with mDIA you also have the challenge of mixing different cell types together like in TMT.

000

Christopher Rose @cmichaelrose.bsky.social · 27/10/2023

Yeah Astral would be good for this. NeuCode would be an attractive approach. You would need to make the tag version of the NeuCode amino acids. We had those at some point in the Coon Lab but I don’t think it was commercialized.

000

Christopher Rose @cmichaelrose.bsky.social · 27/10/2023

You could use NeuCode type of isotopolouges, but that requires very high resolution MS1 scans - which would take longer to acquire. 6-plex needed 1 million resolution. 4-plex needed 500 thousand resolution. So 0.5 to 1 second per MS1 in the Orbitrap.

110

Christopher Rose @cmichaelrose.bsky.social · 27/10/2023

Proteome depth will be a challenge because of this. Also, you adding back the complexity of labeling that is a major downside of multiplexed single cell (assuming this is the context). Like I said though, happy to be proven wrong, but I think the max is more realistically < 10 (likely 5 or 6).

020

Christopher Rose @cmichaelrose.bsky.social · 27/10/2023

Always happy to be proven wrong, but I don’t see how 100+ DIA will work. If each peptide was separated by 4 m/z a single peptide would cover a range of 400 m/z in the MS1. Even with fast instruments and narrow windows, you will struggle to sequence more than the most abundant peptides.

330

Christopher Rose @cmichaelrose.bsky.social · 23/10/2023

It is DIA. Targeted experiments (PRM) focus the center of the isolation window on the peptide of interest, use higher injection times, and peptide specific the CE. All of those improve sensitivity and selectivity. DIA is more generalized -isolation and the CE are not tailored to one specific peptide

250

Christopher Rose @cmichaelrose.bsky.social · 19/10/2023

Excited to share the publication of a large collaborative effort at Genentech to discover shared cancer neoantigen neoepitopes. From 47 cancer neoantigens and 15 HLA alleles we discovered 86 neoepitope-HLA pairs, 65 were novel at acceptance - check it out! www.nature.com/articles/s41...

0125

Christopher Rose @cmichaelrose.bsky.social · 17/10/2023

Not entirely unexpected - but still shocking to see! Mass spec folks may not have noticed, but non-mass spec based proteomics techniques are quite popular already and trending up. Thermo clearly saw this.

250

Christopher Rose @cmichaelrose.bsky.social · 15/10/2023

The challenge is always maintaining that trust. If you give back 20 hits and only 1 pans out…they might be slow to trust you in the future. But it all depend son how you approach it with them. It’s great l these discussions are happening! Hopefully you are enjoying your time down on the Cali coast!

030

Christopher Rose @cmichaelrose.bsky.social · 15/10/2023

Yeah, for sure. We tend to just be up front about the confidence of findings - at times we can work w/ collaborators to go into data that may not pass traditional FDR thresholds. But when we do that we are clear we are wading into dangerous territory and hopefully have some biology to justify it.

240

Christopher Rose @cmichaelrose.bsky.social · 15/10/2023

I agree with the last point…within reason. If the false positives are too high you will spend a lot of time and resources following up on false hits. If you are doing the follow up - maybe not a big deal. If others are doing follow up they might not appreciate it.

210

Christopher Rose @cmichaelrose.bsky.social · 12/10/2023

If I only had MS2 I probably would have been exploring DIA earlier than we are now because for a 10 fold ratio - getting a 4 fold ratio with MS2 is a lot more painful for me than getting 8 fold with MS3. I can with with a little ratio compression for the benefits of multiplexing.

110

Christopher Rose @cmichaelrose.bsky.social · 12/10/2023

It does take a while to optimize, once you have it dialed in it works well. I would say that TMT takes more effort on the front end (sample prep and acquisition) but data analysis on the back end is easier (unless you are crossing plexes).

030

Christopher Rose @cmichaelrose.bsky.social · 12/10/2023

You also have to make sure you get enough signal. We probably inject more ions than others but the quant will be better with more ions. MS3 methods became a lot easier with the real time search, it makes up a lot of the speed and allows you to get more accurate quant…wish they would add it to Lumos

010

Christopher Rose @cmichaelrose.bsky.social · 12/10/2023

Even with MS3 you have to be careful about how you make the methods. For example on the Lumos we always set very narrow ITMS2 windows (0.5 m/z) and then expanded to a charge dependent window for MS3. We narrowed the SPS isolation width and excluded certain parts of the MS2 range around the precursor

111

Christopher Rose @cmichaelrose.bsky.social · 12/10/2023

We almost always run MS3…except for single cell where with narrow isolations our single cell MS2 data and MS3 data correlate well, just with more compression in MS2. With the right real time search there isn’t much reason to do MS2. If you only have an Exploris, then yeah I would be careful.

030

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

We don’t optimize TMT kits as we should. I also haven’t done - rigorous cost comparison. If I did I would also need to factor in the costs of switching methods - retraining and added costs of data analysis. TMT data is very easy to analyze. Most of our workflows are are automated at this point

120

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

I have been looking at DIA papers and data and it appears common to have the first bar of identified proteins and a second bar of quantified proteins, but there doesn’t seem to be consensus on how that second bar is generated. Some folks use CV but that is potentially impacted by chemical noise.

100

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

The hardest thing to wrap my head around in DIA is when did you actually “quantify” something. For TMT it is a simple signal threshold for reporter ions. I haven’t seen data yet that makes me believe for 18 samples you can reach the same depth as TMT with DIA - but maybe I don’t know where to look.

340

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

Agreed! With the current trajectory it is clear that TMT will be losing out to some applications in the near future (although not completely gone). Hopefully the vendors see that too and the competition will drive some innovation on their part.

020

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

I still like what we can do with real-time search today. We have our own IAPI algorithms with custom protein closeout (limiting peptides per protein across runs), protein prioritization (e.g. preferentially sequencing mitochondrial proteins), and we have a preprint out on fixing hyperplexing.

130

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

It’s expensive…can’t argue with that. It is a great tool for the right situations, but I completely understand if some folks are priced out and cost is a non starter. Label free DDA and DIA is going to be hard to beat on sample prep cost.

230

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

Yeah, if you aren’t doing TMT regularly, I would’ve jump into doing 200 samples. We are experienced with TMT and have had troubles scaling up in the hundreds range.

000

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

I think with how fast DIA is now, throughput for TMT is not the greatest selling point. For us it is data depth, completeness, and ease of data analysis (when not crossing plexes). For 18 samples we would run 24 fractions for 90 mins each. So 2 hours per proteome.

270

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

Once you start trying to bridge plexes I would be careful - if you aren’t that experienced in TMT you might want to avoid that all together - as others have mentioned it can be difficult to collect such data in the correct way and then also the analysis is more complicated.

030

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

In these situations TMT give great depth - nowadays 10,000 proteins with real-time search methods we implement. Also the quantitation is simple to digest and allows you to quantify peptides across all samples easily. There is the downside of ratio compression, but we mitigate it as much as we can.

120

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

We use TMT because most of our experiments are small (18-30 samples) and focused on a particular biological question. Often once we do the experiment it prompts a different question that requires a different experimental design. We don’t typically span quantitation across hundreds of samples.

110

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

Also, while we have done mostly TMT due to my background we will certainly be doing more DIA in the years to come because it is makes more sense for the types of questions that are now being asked by our collaborators. Not to mention we have a group leader and DIA expert, Alessandro Ori, who joined

130

Christopher Rose @cmichaelrose.bsky.social · 11/10/2023

Nothing wrong with being both pro-TMT and pro-DIA. I tell folks when they join the department, we just need to use the best technology to answer the biological question. That can be freeing when many academic groups are locked into one approach (TMT or DIA).

260

Christopher Rose @cmichaelrose.bsky.social · 04/10/2023

I was wondering where you were at! Good group of folks here. Good has also been great…hopefully it is as good for you tomorrow!

100

Christopher Rose @cmichaelrose.bsky.social · 28/09/2023

Also something to keep in mind, if for whatever reason you do iodoTMT, you don’t want to check this box. Because b and y ions can have a Cys with a quantitative label in that situation. Checking the box would ditch y ions that end in R but also have C.

100

Christopher Rose @cmichaelrose.bsky.social · 23/09/2023

Hey Amanda, hope you are enjoying Boston and the Gygi lab. Maybe we’ll catch up at a conference sometime.

120

Christopher Rose @cmichaelrose.bsky.social · 23/09/2023

Thanks, good to know. We will be on the lookout.

010

Christopher Rose @cmichaelrose.bsky.social · 23/09/2023

Yeah, this is going to be a big consideration for us going forward. We liked DIANN because of its ability to scale. I guess we will see if they come out with a commercial offering for us non-academics or if we end up using the older versions or something else entirely.

120

Christopher Rose @cmichaelrose.bsky.social · 22/09/2023

It is a lot of filters. The cool thing a building this complicated method was easy, the first version was up and running in a day or two…then spent weeks optimizing and troubleshooting. 😅 It was nice to dig into the data and figure out why hyperplexing tended to underwhelm when I tried it previously

010