Jyunmi

They only allowed themselves 15 AI designs per target.

And even on that limit, they ended up with very tight binders for six of the nine targets.

you will sometimes see this described as nanomolar affinity.

And all that means is the binder still grabs the target very strongly, even when the binder is present in tiny concentration.

Tiny here means billionths of a mole per liter.

Now, this is not just AI juicing pretty molecules on the screen.

This is suggesting the models, labs built them, and quite a few worked well in the tests.

So let's take a look at Boltzgen under the hood.

Boltzgen is an L-atom model, which means it keeps track of where each atom sits in the 3D space, not just a rough cartoon of a protein.

It learns by doing several related tasks at once.

One, predicting how proteins fold from their sequences, filling in missing pieces of protein structures so it has to guess realistic shapes, and then designing binders that fit onto known targets.

By juggling all those jobs, it picks up habits from real chemistry and physics.

It learns, for example, that certain shapes are stable or that some patterns of changes in atoms make good binding surfaces.

The team also uses filters after the generation.

They throw out designs that look like they won't fold or they make ugly clashes between atoms or that they don't seem to interact well with the target.

An important detail is they release the code publicly.

So in principle, any lab with enough computing power and structural data can try similar workflows.

Now,

This is all still very early.

So there's a few things that we have to pay attention to.