Joshua Meier Profile picture
Jan 10 10 tweets 5 min read
📢 Our new #manuscript shows how zero-shot Generative AI can create de novo antibodies from scratch.

🏅Hundreds of antibodies are created zero-shot and validated in the wet-lab for the first time ever.

Paper: biorxiv.org/content/10.110…
Data: github.com/AbSciBio/unloc… Image
Our zero-shot generative AI framework. The model is programmed with a target antigen structure and a chosen antibody scaffold sequence.

Antibodies are then generated de novo.

Note: all proteins binding the target (or its homologs) were removed from the training set. Image
We validate generated antibodies experimentally in the wet-lab.

Our platform is extremely high throughput. We can test ~3 million unique AI-generated designs each week.

A cycle of [AI ➡️ lab ➡️ data] takes just 6 weeks. Image
Example AI-generated cancer drug leads. Our model generates antibodies with stronger affinities than a highly optimized therapeutic drug.

The binders come straight out of the model.

In some cases, >90% of the CDR3 region has changed. Image
Generative models have a tendency to memorize the training set. However, our model generates HCDR3s that are very different than those observed during training (on average, 5/13 positions are changed).

The sequences also differ from known antibodies in massive databases. Image
We predict structures of the discovered antibodies, finding that their conformations are highly flexible.

However, the model still learns to place key residues in appropriate positions, likely mediating binding against the epitope. Image
We're confident in our experimental platform, which can now screen millions of unique AI-generated designs in a few weeks.

To that end, we've opened the data behind this manuscript, releasing the full list of sequences and their measured affinities.

Enjoy! 🤓 Image
We're just getting started. De novo antibody design will unlock a new generation of better therapeutics, with higher probability of success.

We are leveraging this tech to get antibodies into the clinic in just 18-24 months, down from six years.
Special thanks to the dream team for pulling this off 🌟

Not to mention the 200+ unlimiters working to make this a reality.

We have a dozen open roles for AI scientists. Come join us in transforming drug creation w/ Generative AI 🚀 Image
Great coverage this morning on our advancements from @ForbesTech & @VentureBeat:

forbes.com/sites/johncumb…

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Joshua Meier

Joshua Meier Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @joshim5

Sep 2, 2020
Excited to share an update to our work on evolutionary-scale modeling (ESM)! Over the past year, we rewrote our paper with better pretraining and downstream models, leading to state-of-the-art results across multiple benchmarks. (1/8)
biorxiv.org/content/10.110…
Last year, we showed that Transformer language models learn intrinsic properties of proteins from sequences. But on quantitative benchmarks, these models did not improve over alignment-based methods, as shown by @roshan_m_rao, et al in TAPE.😵(2/8)
In this revision, we discovered a major advantage from data diversity. On the TAPE benchmark, our new models now perform similarly to or better than alignment-based methods. 🎉 (3/8)
Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

:(