257 molecules are in on STAT6.
we start validation now. every one gets bought and run on AS-MS against real stat6 protein. not a subset, all of them. binder or non-binder comes back with the measurement behind it, and whoever picked it owns the result.
still open if you want in. 5,000 wallet credits a submission, $1,000 per qualified binder.
omtx.ai/discovery/challenges…
agentic small molecule discovery, from whatever agent you already use.
npx skills add omtx-ai/omtx
that gives claude code, codex or cursor the om mcp and the discovery skill. ask it to score molecules against STAT6 and submit the ones it likes.
we buy every accepted molecule and run it through a mass spec against real protein. binder or non-binder comes back, and you own it.
$1,000 for every Qualified Binder. 25 slots open.
omtx.ai/discovery/challenges…
this is a real discovery program, end to end, from claude code.
fold STAT6 with boltz-2, chai-1 and neuralplexer. score 6.5 million purchasable molecules with LULA. submit the ones your agent believes in.
then we buy those exact molecules and run them through a mass spec against real protein.
you own every result. $1,000 for every confirmed binder.
no wetlab of your own.
omtx.ai/discovery/challenges…
You own your Result Data. Private by default; publication is optional.
If Om determines and selects it as a Qualified Binder within the current cap, Om licenses the data and pays $1,000.
Start:
omtx.ai/discovery/challenges…
Terms apply.
🧵 [4/4]
Why STAT6?
IL-4 and IL-13 signal through STAT6, a central pathway in Type 2 inflammation and diseases including asthma, atopic dermatitis, and eosinophilic esophagitis.
Better experimental binding data can improve discovery models.
🧵 [3/4]
How it works:
1. Use LULA to score molecules.
2. Select a molecule.
3. Submit it for 5,000 Wallet Credits.
4. Om procures and experimentally tests it.
5. Receive your Result Data.
There is no per-participant submission limit.
🧵 [2/4]
LULA-2 Academic Benchmark!
at @decodediscovery we've screened ~8% of the human proteome across 300M molecules each, trained LULA-2 with great performance on internal and public benchmarks. We want to see how well it works for the rest of the community.
here's what we'll do -- we'll use LULA-2 to screen millions of molecules predicted to bind your target of interest we can source in under a week, then we'll test up ~6K molecules in the lab, and we'll give you all the validation data.
we got a ton of interest so we're opening this up more broadly.
we're offering this for $5K per protein, if you supply the protein (~200 ug).
you get to see all the data.
DM if interested!
Attention is all you need small molecule and protein binding edition for structural contacts.
Here’s what lots of data gets you at 100x higher throughput and 100x lower cost than structure based models.
Prospective structures and contact prediction for other proteins in the proteome? DM to learn more.
as we gear up to broadly release LULA-1 and LULA-1.5, one of the more interesting things we've seen is we can leverage the learned attention between ligand and protein to rank residues that show high correlation with ligand binding -- strictly from ligand binding data only, no structural data!
when we convert this to structures with Chai-1 and other models to seed the pocket and ligand interactions, we get high correlation with known PDB structures.
500M data points, 8% of the human proteome, 300M molecules per protein screened in the lab.
DM me for early access.








