matteomics.bsky.social
Measurements want to be accurate;
Experiments want to be elegant;
Data wants to be beautiful and Data wants to be free
#proteomics
#rstats
46 posts
95 followers
370 following
Regular Contributor
Active Commenter
comment in response to
post
The cost saving implications of no samples per condition are a game changer.
comment in response to
post
That is a good point. Personally never occured to me that NCBI and refseq could ever actually be at risk. Possibly the EBI should look at complete replication for the sake of redundancy.
Sad US researchers wanting to replicate work in future need to consider risk of US databases disappearing.
comment in response to
post
So EMBL and www.ebi.ac.uk/uniprot/index are still there! Maybe time to sample the European tools and data sets! Personally I find the uniprot website richer and better designed than NCBI.
comment in response to
post
I usually re-heat it in the microwave, and forget to take it out again.
comment in response to
post
Took the kids for their COVID vaccines. NHS nurses were perfect, comforting, provided colouring books, distractions, a sweet drink.
On the way out the kids found a feedback terminal. Banged out a sad-face each because it was a vaccination.
I hope no-one's performance review depended on that data.
comment in response to
post
It was on an issue/help request page for ellmer somewhere.
comment in response to
post
If you set option thorugh:
options(ellmer_timeout_s = <ms>)
It alters the time out. Should probably go in a .Rprofile file if used often.
See below, I asked qwen2.5-coder about The Lord of the Rings just to test the longer timeout. It's wrong, but close, a bit like the code it produces...
comment in response to
post
I have lots of RAM and a GPU for mass spec data processing. Gander seems to do better in files with more code. I wonder if it's getting context for what's being asked from the whole file.
Had to set an environment variable to a longer timeout.
comment in response to
post
Just my notes on using gander in R with ollama qwen2.5-coder:32b running localy. Not convinced it's going to be terribly helpful. Made simple mistake in example code.
comment in response to
post
I did get it working. Running it locally is frustratingly slow compered to speed I expect of a server. But it's still much faster than writting code manually and no data goes externally which I like.
comment in response to
post
Have been playing with local llm today. Not looking great so far. Have been trying deepseek-coder-v2 to start with. Happy to report back.
comment in response to
post
Impartiality was an effective unthreatening posture to influence policy in normal times. But what if the value of education, or science influencing policy, themselves become febrile political questions? If the science community won't turn up to defend evidence based policy, who will?
comment in response to
post
An engineer once explained to me that an efficient system always has need available when a resource becomes free, and an effective system always has a resource available when a need arises. Mind was blown.
comment in response to
post
I wonder if future AI tools will be developed to identify and pull out papers written by real people with interesting well expressed ideas. Like craft beer or artisan cheese, those are the papers that will have value.
comment in response to
post
Love that list. The question of the age is how such mild mannered scholars can find a way to effectively oppose the gathering dark. And without losing the virtues in that list.
comment in response to
post
Well I like Zotero. It has an extension that works with #Notion and another that works well with #quarto and #Rstudio if either of those things are useful. Syncs across devices. Will store documents for a fee. www.zotero.org
comment in response to
post
So, on one view those p-values are just what that approach produces. Intuitively it just feels wrong because surely all p-values should be different.
Maybe swap the BH out for Storey's q-value to adjust raw p-values? Perhaps less conservative, can be odd with high density p-values near 1.
comment in response to
post
Yeah. I think old money was probably easier for day to day spending and wages than decimal currency. But a nightmare for spreadsheets, interest, inflation and high finance. We think in spreadsheets now. It's hard to imagine a different world.
comment in response to
post
A lot of it is secretly base 12 and can be divided 1/2 1/3 or 1/4 into whole units. That's a really useful feature in a semi-numerate culture.
comment in response to
post
Lastly I compared protein quantification obtained from DIA-NNv2.0 and DIA-NN v1.9. Plot shows y=y line in black, regression line in blue, data points in red with high transparency. Quantification is fairly tight. DIA-NNv2.0 underestimates wrt v1.9.
comment in response to
post
Once data is in it needs some tweaks to get the columns named in the way MSstats expects. Fragment data is now included in separate columns and has to be concatenated into a single column for MSstas.
comment in response to
post
Getting .parquet file into #rstats was easy. There’s a package {nanoparquet} on cran. Function read_parquet() just pulls in the table. Actually all columns come in with correct type, no more sequences coming in as factors!
comment in response to
post
My existing pipeline works through {MSstats} in #rstats #rstudiofrom @bioconductor.bsky.social which needs precursor and product ions to quantify proteins across conditions. Running DIANN v2.0 through command line requires the --export-quant switch to include fragments in the .parquet output.
comment in response to
post
Screen shot of viewer
comment in response to
post
Also, you could imagine an AI arms race in which businesses buy AI systems tuned to turn customer AI calls into sales, and customers buy AI systems tuned to find the best deal. Both must constantly update to avoid missed sales or over paying.
comment in response to
post
Oh. Cool. Thank you.
comment in response to
post
I have never seen a sequence with R group illustration instead of AA letters before. How do you get that?
comment in response to
post
Cool. I will train myself to habitually use the base R version. Thanks for this.
comment in response to
post
Cool. Thanks.
comment in response to
post
Possibly you've essentially answered that already in your other recent post here: bsky.app/profile/jben...
comment in response to
post
I'm only just getting into using pipes routinely. Is one of these flavours more correct, or more likely to become the standard? and so worth habitually using from the beginning.
comment in response to
post
My first time was a Thai lunch place in Manchester. Kimchee fried rice with egg on top, sriracha sauce and ginger beer. Life changing.