What does it do?

I am so excited to share our new findings with you! We provide the structural evidence for a direct protein-to-DNA information pathway, showing how a bacterial enzyme 'reads' its own structure to 'write' DNA. www.science.org/doi/10.1126/...

Seize the moment and redefine how research money is distributed in a fair way in an age where the quality of a written proposal is increasingly disconnected from the quality of the research. Too difficult? Lottery is pretty good

We got internal info today that hints at what at least ERC thinks is a good (=their only) idea to counter the proposal flood: vastly stricter resubmission criteria! Retroactively even! 🤬

Makin a bunch more stability updates and bug fixes for my TUI plasmid editor. Currently workin on a pLannotate wrapper module so you can annotate plasmids imported where the author did not fully annotate all the features. I'll eventually try to pull plasmidsaurus seq results too

I also don't think that reasonable, though in my scenario, the OSS author's thought is at least not unreasonable. It did, however, become defunct

Letter of the law vs spirit of the law?
Can't you imagine that some OS their code because they think "if some person is interested enough in my software to work through it to the point they understand & effectively modify it, they should be able to do so w/o restrictions"

Sequence Display generates large-scale protein sequence–activity datasets by simultaneously reading out the sequence of interest and a downstream editing efficiency - perfect for data hungry ML methods!

Happy to have played a small part in this!

an xray density map depicting blue meshed density where an atomic model has been built.

Some people cleave tags religiously. This person didnt and made me solve their perfectly structured Histag. Probably also helped in packing. Might have to do experiments without it to show its not an artifact though.

Team cleave or not cleave?

Torn between amusement and frustration that @alignbio.bsky.social cancels PETase tournament halfway through. Cryptic mail suggests they hugely underestimated how difficult measuring enzyme activity in high throughput really is. Better pulling the plug than lousy execution, but.. poor planning?

But the real question is - what do we actually *want* models to produce? Is the model creating 92% designable structures great and the one with 25% crap? We argue that it is actually the metric that is flawed: you may *want* (some) unstructuredness, in which case measuring rmsd is just senseless

Counterintuitive because discrepancy so high? Not an expert in diffusion models, but I guess some models /modes may implicitly learn to generate what we consider "designable" structures simply by avoiding the unstructured "mess" seen in so many tail-to-tail AFDB structures. 1/2

On paper, but the reality might differ from a bit to a lot
Anyway, it's an individual circumstance and thus should be an individual decision if one can "afford" to take that risk
Thus my skepticism toward framing it as a matter of "now I'm wise, before i was dumb (implied: others still dumb)

The conundrum (that journals exploit) remains, as ECR, how can / why would you take the risk?
"Sorry folks, no tenure for me, the lab closes in the fall, guess I should have tried Nature XYZ for our big story last year after all"
Funders/promotion committees have to change first, scientists follow

Wiser or just comfortably cushioned because now you already have those CSN papers and made a name of yourself?

Y'all have been using chatgpt wrong. Instead of carefully avoiding outsourcing your thinking and only coyly resorting to it to fix your grammar, turns out you just had to let it do your entire job and then go brag about it to get your Nature paper

Use the AI-Powered 3D Structure Similarity Search This new feature delivers faster and more scalable structural comparisons across experimentally determined and predicted Computed Structure Models (CSMs).

Use the AI-Powered 3D Structure Similarity Search
This new RCSB PDB feature delivers faster and more scalable structural comparisons across experimentally determined and predicted Computed Structure Models (CSMs).

sorry i never responded to your email, i didn't want to and then i forgot

Stryer

Consider yourself lucky if you haven't developed "token anxiety" yet.

Simulated Annealing — interactive explainer Step through the process of Simulated Annealing: initialization, temperature schedule, neighbor selection, and acceptance criteria.

More fun with #ClaudeAI's ability to build interactive demos. This time, I described a Simulated Annealing/MH sampler demo, and it gave me this nice tool to embed on my Canvas LMS pages.
tpavlic.github.io/asu-bioinspi...

We're also releasing a massive self-distillation set, a key ingredient in training AF3, comprising millions of diverse MSAs and structures. We estimate its cost to near $20M, representing perhaps the largest compute investment by an academic effort for a biological dataset. 6/9

...A small Dutch town decided – for lolz – to build all those fictional bridges. As canal crossings.
So you can visit Spijkenisse, nr Rotterdam and see the real (fake) bridges. Made of concrete. Deeply weird and silly. But fun. 2/2 🧵

Good news, protein engineering researchers, mutation effect predictions remain an unsolved problem.

Ig Nobels to move awards to Europe due to concern over US travel visas Scientific awards – which honor research that makes people laugh and then think – to move away from ‘unsafe’ US

BREAKING: For the first time in 35 years the Ig Nobel ceremony will move to Europe. "There are no immediate plans to return the ceremony to the US."
www.theguardian.com/science/2026...

Food for thought for conference steering committees.

Let’s work together with Slack Connect Accept this invitation and our teams will be able to send messages, share files, and work together right in Slack. Depending on your settings, your admins may need to approve your request first.

At OpenFold, we want to make it easier for the community to connect with us for OpenFold3 discussions and support. Join us via Slack Connect using the invitation below!

join.slack.com/share/enQtMT...

guess they're just too busy playing the other game

Finding one mutation that improves a protein is hard. Finding five that work together is exponentially harder.

Today in @science.org, Hsu and Konermann labs present MULTI-evolve, a lab-in-the-loop framework that does it in just one machine learning-guided round.

Did you go through the issues on GitHub already? This was a recurrent complaint and I think some may have provided files / details

Stupid name for a social network anyway.
Hey great paper by the way 😊