We figured out flow matching over states that change dimension. With "Branching Flows", the model decides how big things must be! This works wherever flow matching works, with discrete, continuous, and manifold states. We think this will unlock some genuinely new capabilities.


For example, here is what "binder design" looks like when you start from a single amino acid.

Real-time data update: we've got our first BA.2.86 (see red arrow) neutralization results. Short 🧵with much credit to @DannySheward and the rest of the team. Our initial data takes two cohorts of blood donors from Stockholm: one from before XBB arrived in late 2022, and one from last week. Exposure histories (ie. vaccination and infection) are exceptionally diverse at this point, so we prefer a random slice with well-defined timing.

BA.2.75.2 exhibits more extreme antibody escape than any variant we've seen so far. A short 🧵 As with our previous work, this effort was led by @DannySheward, with @PeacockFlu helping us accelerate things.

We recently discovered that a “public” class of antibodies (one that most people should be able to make) can mature to cross-neutralize Omicron, and all other SARS-CoV-2 variants we tested: biorxiv.org/content/10.110… This story is the result of a three-lab collaboration with the labs of @NillaKH and @BMartinHallberg (who also made the structure figures I'm posting here), with @DannySheward, @Pradee05, and @das_hrishikesh1.

Another Omicron neutralization thread, this time about monoclonal antibodies. Once again, massive cred to @DannySheward, and everyone else from the previous data release. This time, we also need to especially thank @ReddyLab_ETHZ (1/n) It turns out that panels of relevant monoclonal antibodies are non-trivial to find in a hurry, and @roy_ehling and @saireddy911were happy to provide these ASAP. Before results, some caveats: (2/n)

Omicron antibody neutralization thread, with preliminary results, and substantial caveats. Most credit to @DannySheward. We’ve been racing to generate neutralization data as fast as possible, and our first results were read this afternoon (1/n). Major NB: this is preliminary data. Given the circumstances, however, we feel it critical to release data right away, and explain current caveats that we can later iron out and investigate. Link here: tinyurl.com/ycx4x4d4 (2/n).