iiinova — where your digital self keeps evolving

Be careful! Those peptides that did not fragment could go off at any time. Magnus probably knows which amino acids are the sharpest.

010

Phil W @pwilmarth.bsky.social · 09/10/2026

Have you rewritten Sage in Assembly or CBasic for CP/M and Z80 CPUs (en.wikipedia.org/wiki/Zilog_Z80)? How big a fragment ion index can you fit in 64 KB of memory? How are you distributing it? 8 inch floppies or 5.25 inch?

en.wikipedia.org

Zilog Z80 - Wikipedia

220

Phil W @pwilmarth.bsky.social · 07/10/2026

The end days are so difficult to get through. Our condolences for your loss.

010

Phil W @pwilmarth.bsky.social · 07/10/2026

I think the point is less about gate keeping and more about slamming the door shut.

010

Phil W @pwilmarth.bsky.social · 07/10/2026

AI is societal mass suicide. We have only seen the testing of a small quantity of shit hitting a small fan. No one will be ready for the full on large-scale deployment.

static.klipy.com

That's Nice Honey: Drinking The Kool-Aid Meme

ALT: That's Nice Honey: Drinking The Kool-Aid Meme

010

Phil W @pwilmarth.bsky.social · 06/10/2026

The committee is still amazing. You would think they would be running out of old white males that they can give awards to instead of women, but no...

000

Phil W @pwilmarth.bsky.social · 03/10/2026

I propose we call this scientific revulsion "last-generation proteomics". One old white scientist can do the work of thousands of young scientists depriving them of scientific careers thereby breaking the generations long cycle of passing down knowledge. Go Team!

010

Phil W @pwilmarth.bsky.social · 01/10/2026

Linus Pauling answered the question of how do you have so many good ideas with you need to have lots of ideas and get rid of the bad ones. Even Pauling had more bad ideas than good ideas. I want barriers to implementation of bad ideas! Science is full of bad ideas - they are called conferences...

130

Phil W @pwilmarth.bsky.social · 01/10/2026

Unfortunately, there won't be any liberal arts colleges left in the AI (or SHIT) future and probably no higher education, period. Unless Billionaires want their kids out of the houses for 4 years to party and find spouses. Oh, and to bet on college football/basketball (for men, of course).

static.klipy.com

2001: A Space Odyssey Monolith and Apes

ALT: 2001: A Space Odyssey Monolith and Apes

110

Phil W @pwilmarth.bsky.social · 30/09/2026

I am pretty sure every Walmart greeter was not too worried about ever being a Walmart greeter...

120

Phil W @pwilmarth.bsky.social · 30/09/2026

Ha-ha. The AI has probably already rewritten Sage to meet its requirements and made you irrelevant. Time to use up the vacation and sick leave before retiring...

330

Phil W @pwilmarth.bsky.social · 30/09/2026

I think the new name for AI is not quite right yet. It should really be Super *Human* Intelligence *Techbro-ology* (SHIT). Grab the Wellies, time to wade through some chat... Grab a new diaper, some more *agents* have leaked out... It is time for Claudia do my homework again...

020

Phil W @pwilmarth.bsky.social · 30/09/2026

Chip (my cat) says this is what happens when you let dog people run the country...

000

Phil W @pwilmarth.bsky.social · 29/09/2026

I think over-reliance on algorithms (without checking or validating for every single case) is one of the major issues in proteomics data analysis. My approach visualizes target/decoy for every dataset so that delta masses, PTMs, sparse data categories, etc. can be checked by an actual human.

021

Phil W @pwilmarth.bsky.social · 27/09/2026

You can try different normalizations and see how results are affected. Focus on things that you think are strong negative controls. I have some thoughts on prepping data for quant analysis at github.com/pwilmart/qua.... There is also a discussion of QC metrics.

github.com

GitHub - pwilmart/quantitative_proteomics_data_cleaning: A discussion of basic data cleaning concepts for quantitative proteomics data.

A discussion of basic data cleaning concepts for quantitative proteomics data. - pwilmart/quantitative_proteomics_data_cleaning

010

Phil W @pwilmarth.bsky.social · 26/09/2026

It is basically just scaling to match the median values. The trimming of some of the data just makes it a bit more robust that the median is more representative of the unchanged distribution.

100

Phil W @pwilmarth.bsky.social · 26/09/2026

They are clearly getting "market growth" confused with "marketing department growth". Maybe, with some parsimony analysis, they can reduce the FPP (false proteomics people) to manageable levels.

010

Phil W @pwilmarth.bsky.social · 26/09/2026

The trimmed mean of M-values normalization was designed for that: link.springer.com/article/10.1... It's part of Bioconductor edgeR package. It does single factor scaling per sample. Scaling factors are used by edgeR scripts and the actual normalized values aren't computed. Divide by the factors.

link.springer.com

A scaling normalization method for differential expression analysis of RNA-seq data - Genome Biology

The fine detail provided by sequencing-based transcriptome surveys suggests that RNA-seq is likely to become the platform of choice for interrogating steady state RNA. In order to discover biologicall...

110

Phil W @pwilmarth.bsky.social · 25/09/2026

A lot of the top labs have papers I would not want my name on...

020

Phil W @pwilmarth.bsky.social · 23/09/2026

If it has one tail, you can eat it for supper. If it has two tails, that came from near the Nuclear power plant and you shouldn't eat that...

010

Phil W @pwilmarth.bsky.social · 22/09/2026

Morally, I have to look at the whole technology and its effect on society. No AI for me. I am sure there are some really cool aspects to assault rifles, but I can't get past all the killing and murdering. Keep your data center off of my front lawn!

static.klipy.com

Old Lady Yells 'Get Off My Lawn!'

ALT: Old Lady Yells 'Get Off My Lawn!'

130

Phil W @pwilmarth.bsky.social · 21/09/2026

If AI eliminates that process, it eliminates most learning and all expertise. You will lose skills, knowledge, and insight by using AI. AI will be worse than the Vitamin D water in Silo. AI will do to human brains what Xenos peckii insects do to native paper wasp brains.

110

Phil W @pwilmarth.bsky.social · 21/09/2026

I learn new things with each data analysis. However, where I learned the most was coding my pipeline scripts. You have to figure out and deal with all the edge cases. You have to make compromises and practical choices. You have to evaluate all of that. That process is where I really learned things.

130

Phil W @pwilmarth.bsky.social · 21/09/2026

I think the learning and expertise comes from the journey not the destination. AI takes away the journey. I have done tons of reading on computational proteomics and proteomics tools and I learned a lot from that. I have spend almost 20 years using my analysis tools on a huge number of projects.

350

Phil W @pwilmarth.bsky.social · 20/09/2026

The big AI companies are now doing actual product demonstrations for their biggest potential clients!!! After the tech(just say no)bros have emptied the World's government wallets, they are coming for organized crime's money. Did someone say crypto? Cha-ching!

001

Phil W @pwilmarth.bsky.social · 18/09/2026

One of the cops has a very defensive pose. I can't tell if he is anti toilet paper or pro vandalism.

000

Phil W @pwilmarth.bsky.social · 17/09/2026

See Dick load his gel. See Jane load her gel. See spots run. Run, spots, run!

110

Phil W @pwilmarth.bsky.social · 17/09/2026

The key is wrong, too. Blue circles are fact-based courses, white circles are opinion-based courses (discussions, etc.), and the red circles are just all lies...

000

Phil W @pwilmarth.bsky.social · 14/09/2026

Whoa there. We already have too-many-to-count enrichment analysis tools to tell us deep, insightful things like, "inflammation is an important process in many inflammatory diseases" and "cellular component".

120

Phil W @pwilmarth.bsky.social · 14/09/2026

Elementary school science fair projects will never be the same!

130

Phil W @pwilmarth.bsky.social · 12/09/2026

The science is in the selection of samples, the experimental design, and the quality of the sample prep. The LC, MS, and software can screw that up, but it has never and will never be able fix the most important parts of proteomics experiments. The "advances" are always in the wrong places.

172

Phil W @pwilmarth.bsky.social · 10/09/2026

Wow. That is even more teeth than Enric Mas's smile!

010

Phil W @pwilmarth.bsky.social · 10/09/2026

Nano particle enrichment techniques are a rather expensive and complicated way to do sample and sample processing QC. I guess it is better than nothing...

010

Phil W @pwilmarth.bsky.social · 10/09/2026

If the dems wanted to make some extra cash for the midterms, they could print toilet paper with facial images on each square. I know someone who likes their image on everything...

000

Phil W @pwilmarth.bsky.social · 10/09/2026

What are the odds that AI decides to kill itself? It seems most recent techbro products increase suicide rates...

010

Phil W @pwilmarth.bsky.social · 07/09/2026

I did not realize that life sciences were not getting shitty fast enough (it seemed pretty fast to me). With some small multiple of the annual electrical and water use of the State of Washington, I have no doubt that Anthrodick can use LLMs to remedy that in less than 100 days...

100

Phil W @pwilmarth.bsky.social · 05/09/2026

And they need to use the 2 Scientists per Prize rule. Single scientist per prize awards are too noisy unless they are manually validated.

100

Phil W @pwilmarth.bsky.social · 05/09/2026

There will be about 100 small awards made, one of them will likely be wrong, and no one will know who actually got the award until the winner is inferred based on a set of ad hoc rules published in MCP in 2004-ish.

110

Phil W @pwilmarth.bsky.social · 30/08/2026

Cleaning up the scientific record is a job with even less financial support and status than cleaning lab glassware. The AI gardenhose of crappola has already made this orders of magnitude more impossible. Just wait until the AI shit spigot is the largest firehose ever seen in the history of brotech.

010

Phil W @pwilmarth.bsky.social · 27/08/2026

Who really cares what they call Lake Natalie? It will always be my favorite of the "tradlakes"...

100

Phil W @pwilmarth.bsky.social · 27/08/2026

MUSC Impossible: Ghost Protocol. Starring Tom Cruise as Dr Fauci...

010

Phil W @pwilmarth.bsky.social · 26/08/2026

Time to put on the "paper reading" Wellies...

110

Phil W @pwilmarth.bsky.social · 25/08/2026

The N-term of peptide are fun: ragged ends, N-term Q or E can lose ammonia or water, C+57 at N-term forms rings and loses ammonia, etc. Do those happen during sample prep or in the ion source? Are any biological? Ha-ha. It never ends. One-in-a-million not so rare when there are millions of MS2.

120

Phil W @pwilmarth.bsky.social · 25/08/2026

Deamidation may seem unique. But what about acetylation and carbamylation? Both PTMs are in older lenses. Carbamylation can be caused by urea preps. Carbamylation has prominent neutral loss, acetylation doesn't. Both cause missed cleavages and can't be at C-terminus except possible protein C-term.

120

Phil W @pwilmarth.bsky.social · 21/08/2026

From what I have seen, graduate students have a real talent for finding the worst datasets in PRIDE for method testing and development. Since LLM AI is as smart as a graduate student, does that mean AI is real good at finding bad datasets???

000

Phil W @pwilmarth.bsky.social · 21/08/2026

Hard problem. Multiple PTMs per peptide look noisy. Noise likely not random, so decoys underestimate errors (like DIA). There are non-shifted fragment ions with any PTM that contribute to match score. That might let incorrect target matches with PTMs score higher than modified decoys (target bias).

030