learning about agents x ml x bio @generate_biomed | prev @ginkgo @brownuniversity
Boston, MA
Joined July 2020
- Tweets419
- Following493
- Followers2.4K
- Likes2.6K
This plot isn't normalized by expression.
That's great point because for a long time we understood Happy Stoat's advantage to be primarily expression-driven, i.e. there isn't a strong potency advantage per surface CAR.
Some more recent data has changed our perspective somewhat but there's a lot more to understand. For example, the gap between Happy Stoat and the 2nd gen clinical reference can be binder-dependent. With an anti-CD20 binder, we saw a more pronounced potency gap which holds after normalizing by expression.
Yes definitely. Reduced payload size for in vivo delivery was our initial motivation.
But to our surprise, we discovered many other benefits with the compressed domain, such as increased expression, killing, and persistence.
Liam Bai retweeted
The "Age of Miracles" era:
I took a look inside Anthropic's latest progress in the life sciences, including working with a ton of biopharma partners and a new acquisition disclosure
All coming before what is likely the business world's largest IPO:
endpoints.news/anthropic-pit…
Great questions.
We have tested the lead construct across many binders/targets, while the data in this tweet thread are primarily from FMC63. More details in our preprint: generatebiomedicines.com/wp-…
With off-target was referring to tonic/background activation in antigen-negative cell lines.
We don't have data on immune synapse size yet but that would be really cool!
Previously the team focused more on binder design. With binders of different lengths (including e.g. miniproteins), we haven't seen this pronounced an effect on expression.
Although it's hard to disentangle the human vs. agent contribution in this process, I believe that a human expert could have never come up with some of these designs alone.
Overall, I see the role of agents here as enabling something that wasn't possible before, rather than speeding up an existing process.
Second, I see our designs as really quite unconventional, straying far from consensus.
For example, a new hybrid ITAM + TRAF motif.
nitter.cf/liambai21/status/21152…
First, we stand on the shoulder of giants.
This is even more true for agents, who can consume and reason over literature much better than we can.
None of this would have been possible without decades of exceptional research in this area.
This was foundational work that motivated us to go after this problem in the first place.
pmc.ncbi.nlm.nih.gov/article…
We weren't the first ones to go after shrinking the signaling domain, either.
Feucht et al. tried variants with only one ITAM out of the usual three and found killing to be preserved.
Surprisingly, they actually found the CD40 domain that worked best in our format to be inhibitory in theirs.
Goes to show how little we understand this sequence space.
science.org/doi/10.1126/scie…
Daniels et al. screened thousands of costim motifs in this format, and trained ML models on the data.
For context, see this thread on our work and collaboration with @AnthropicAI
nitter.cf/liambai21/status/21152…
What will drug discovery look like when intelligence is abundant?
Our team at @generate_biomed began with this question, and we’re excited to share some results!
We took on in vivo CAR-T and designed a scaffold 70% smaller than ones in the clinic, with 3x the potency 🧵
Great analysis from @owl_posting as always.
Expanding a bit on our work @generate_biomed particularly on the literature, novelty, and creativity 🧵
Liam Bai retweeted
This was a surprise and updated my thinking. Would have expected this level of performance improvement to have come from a fine-tuned protein model and lots of training data specific to the task, not off-the-shelf frontier LLM.
Great work @generate_biomed and @AnthropicAI !
Liam Bai retweeted
Love reading about these projects that would have taken entire PhDs to do previously. We're going to see a lot more elegant, self-contained solutions like this in the coming months...
What will drug discovery look like when intelligence is abundant?
Our team at @generate_biomed began with this question, and we’re excited to share some results!
We took on in vivo CAR-T and designed a scaffold 70% smaller than ones in the clinic, with 3x the potency 🧵
Thanks, Tanishq. Have been following your work and congrats on Holon-S1!
Would love to hear your thoughts on our work with in vivo CAR-T:
nitter.cf/liambai21/status/21152…
What will drug discovery look like when intelligence is abundant?
Our team at @generate_biomed began with this question, and we’re excited to share some results!
We took on in vivo CAR-T and designed a scaffold 70% smaller than ones in the clinic, with 3x the potency 🧵
Liam Bai retweeted
We have barely scratched the surface of ICD design.
Shout out to @liambai21 for leading this work.
What will drug discovery look like when intelligence is abundant?
Our team at @generate_biomed began with this question, and we’re excited to share some results!
We took on in vivo CAR-T and designed a scaffold 70% smaller than ones in the clinic, with 3x the potency 🧵
Liam Bai retweeted
Incredibly cool to see. This + the one-man biotech acquisition by Anthropic are lead indicators of the future
AI will spawn a thousand flowers in biology. “Why not me?”
What will drug discovery look like when intelligence is abundant?
Our team at @generate_biomed began with this question, and we’re excited to share some results!
We took on in vivo CAR-T and designed a scaffold 70% smaller than ones in the clinic, with 3x the potency 🧵
Liam Bai retweeted
A smaller CAR T-cell receptor could make room for a more capable therapy.
In our collaboration with @AnthropicAI, six of our scientists used @claudeai alongside our own models, computational tools, and wet lab data to refactor a CAR T-cell receptor for solid tumors.
The redesigned signaling component was about 70% shorter than a standard design, making room for new functions and improving cancer-cell killing in laboratory tests. Watch how our scientists turned years of work into months, at a fraction of the cost, expanding what we can engineer for patients.
@AnthropicAI's case study → anthropic.com/research/early…
Our preprint → generatebiomedicines.com/pub…
Our blog → generatebiomedicines.com/blo…