Kelvin Lau @klausenhauser on 🙅‍♂️🐘🪡

Look at these colours!

Eating my first non-cherry tomato. Everett’s Rusty Oxheart! Juicy, sweet, weakly acidic. And those colours!

It’s a rare variety but lots of seed vendors

They also don’t want to do it via this traditional evaluate grants give money system.. from friends that have moved to investing into science, they have a more business and applied perspective. They’re investing in science, just not the way academic science is structured and currently functions

I think around 50-60%

I’m literally in a cell culture incubator. I think the first time in my life.. well other than the days before my birth

When the reviewers' identities are revealed

I am super excited to announce that I will be starting my lab at the Department of Pharmacology of the University of Zurich in Switzerland next year!

Gonna sign up to try. Very interesting for my non hardcore (ie scripts+coding) users

Introducing Latent-X — our all-atom frontier AI model for protein binder design.

State-of-the-art lab performance, widely accessible via the Latent Labs Platform.

Free tier: platform.latentlabs.com
Blog: latentlabs.com/latent-x/
Technical report: tinyurl.com/latent-X

Working with @astbury-bsl.bsky.social, we report cryo-EM structures of CD163 in its apo state and in complex with haptoglobin–hemoglobin (HpHb), and propose a framework of how CD163-mediated endocytosis clears different isoforms of HpHb. Please check it out on @plosbiology.org!

That’s also very important. The transparency is better for all. But I do understand concerns that are not directly related to the science but with jobs/careers. Very weird incentives we have built into the system.

Yup in whatever case I think that’s the advantage of preprinting it. Imagine not only the time lost for research and science in general but the impact it has on the careers of these researchers

Potentially. But they preprinted it. So after that I don’t really mind if someone chooses to do peer review or not.

But they only decided to submit pretty much end 2023/beginning 2024… so review process was just a year.. actually not bad for a Nature branded journal then

Disorder meets its match Designed protein pockets recognize intrinsically disordered protein regions

Designed protein pockets recognize intrinsically disordered protein regions @science.org
www.science.org/doi/10.1126/...
www.science.org/doi/10.1126/...
www.statnews.com/2025/07/17/n...

More tightly bound Ni. Similar to competitors like Ni Excel or Ni Advance. Can use super higher levels of chelators and reducing agents. We use it since we find it greatly reduces unspecific binders. Probably because the Ni is more complexed already? Can clean them just by dumping into NaOH

And it’s glorious

The backstory here is that the Squamish people of what is now Vancouver were illegally dispossessed of this land a century ago. They sued, got their land back, and are using their sovereignty to ignore local zoning rules and build 6,000 new homes over the objections of nearby wealthy NIMBYs.

Just cleaning some resin. This batch we started using in April 2025. Cube Biotech can you ship some more over here? ;)

We then fine-tuned the model on a larger scale, to predict forces across 10K solar systems.

We used a symbolic regression to compare the recovered force law to Newton's law.

It not only recovered a nonsensical law—it recovered different laws for different galaxies.

To demonstrate, we fine-tuned the model to predict force vectors on a small dataset of planets in our solar system.

A model that understands Newtonian mechanics should get these. But the transformer struggles.

So so true

Can an AI model predict perfectly and still have a terrible world model?

What would that even mean?

Our new ICML paper (poster tomorrow!) formalizes these questions.

One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵

Village cat meowing begging for pets

Petting village cat

Village cat in ecstasy

Village cat leaving its odour on my bags

Special bonus 3/2 for those #cat lovers

Mountains with snow and a cabin in the distance on a ridge

Cabin up close enclosed in metal

Sunset

Morning sunrise and with a flagpole

Some science talk then off to destroying knee ligaments 1/2

Today in the journal Science: BioEmu from Microsoft Research AI for Science. This generative deep learning method emulates protein equilibrium ensembles – key for understanding protein function at scale. www.science.org/doi/10.1126/...

BioEmu 1.1 can emulate equilibrium distributions of millisecond-timescale MD at many orders of magnitude speedup, bringing GPU-years down to GPU-hours.

The 0737 Saturday train out to the mountains is a great way to bump into familiar faces!

I think I’m addicted growing things. At least they’re not as disposable as Ecoli