62 AI protein and antibody design tools. We read every licence at the source.
Only 43 are a clean "commercial use OK":
- 6 restricted
- 6 non-commercial
- 7 unverified
That gap is why we built a directory. Thread:
@MartinPacesa Single-sequence prediction framed as new is odd, agreed. What would settle it is a held-out set with measured binding, not another benchmark split — if cropping plus single-sequence changes anything, it should show up as a different assay hit rate, not a different DockQ.
@heyshrutimishra Your last line is the whole question. On the Adaptyv round the useful signal wasn't the 354 that bound — it was expression and solubility knocking out candidates before any affinity readout. A structure score never paired with a measured Kd and monomer fraction isn't calibrated.
@curiouswavefn The unglamorous bookkeeping decides this: expression yield and monomer fraction first, then affinity. A real share of in-silico Pareto-front miniprotein designs fail before you reach a Kd — poor expression, or aggregate off SEC. Report the drop-out per stage, not just the hits.
@assemblingfries@KingLabIPD@biorxivpreprint@UWproteindesign Makes sense — at that scale EM is the readout that actually tells you whether it assembled, and SEC can't really resolve it. Thanks for sharing the details, looking forward to the full paper.
@assemblingfries@KingLabIPD@biorxivpreprint@UWproteindesign Thanks — really helpful. Starting from validated oligomers keeps expression off the critical path, so the failures move to assembly. Did SEC alone separate the single-state designs from the aggregators, or did you need a second readout?
5/
Still missing: no SEC monomer percentage, KDs from two technical replicates, and an SPR format with IgG flown over immobilized antigen, so apparent affinity, not monovalent KD. No animal data.
Takeaway: stability moved with affinity, not against it.
https://t.co/U75cxSyiJC
1/
Seq2Study Wet-Lab Check #13
18 of 59 CDR positions were even mutable. 115 design calculations. 6,912 variants, one shot, no iteration rounds.
Best measured result: 50 pM by SPR, over 30-fold better than the parent antibody.
Here is what the lab actually ran.
@sesameandrea That split makes sense: ranking before ordering saves the most, and prioritizing wells for monomer purification is a cheap second readout. If you test it, SEC monomer fraction vs predictor rank per well would show quickly if it helps. Congrats on the preprint.
@johncumbers@JkCla@Ginkgo Novo buying the lab itself, not just samples, is a different scarce-seat bet. Do the Autonomous Labs SKU and the Antibody Developability Consortium assays (Takeda, Argenx, Lundbeck, AbbVie) return numeric retraining-ready columns—expression, SEC, SPR/BLI KD—or only a pass/fail?
@BioAIFrontier ~7,000 variants with SPR gains up to ~30×, and a 10-design screen that improved affinity by an order of magnitude without collapsing onto one winner. What share cleared expression and SEC monomer before that SPR number, and was developability a numeric column next to KD?
@jimmy_build2026 Design throughput is already ahead of assay queues. For a de novo binder campaign, which verification SKU would you buy first on that China CRO layer—expression/SEC monomer, SPR/BLI KD, or developability—and do the columns come back numeric and retraining-ready?
@DrLeucine@adaptyvbio 40 GLP-1R designs with a secret sauce is a bold competition entry. When Adaptyv posts the bench card, will ranking be expression-first, or binding KD (SPR/BLI) among the ones that cleared monomer SEC?
@samsinai Structure is where foundation models clearly win; function is the messy column. On the 50% bet that LLMs beat pure PLMs for sequence design in 3 years: which wet-lab metric decides—binding hit rate, expression yield, or developability?
@Junioryu136689 97.8% stepwise DNA yield and 11k hardware ops are a serious autonomous-lab bar. Can the same multi-agent loop close on protein expression, SEC monomer and SPR without a human in the decision path, or does the assay stack still break the loop?
@BiologyAIDaily A 30.4x lysate fold-change on a second-ranked 5-mutant is the prospective number digests rarely lead with. Was that purified-enzyme kinetics or crude lysate only, and did SEC/expression filters sit in front of the assay?
@WengongJin Recursive policy improvement with wet-lab scoreboards is the right bar. For the 8 selective pMHC binder hits: confirmed by SPR/BLI after expression/SEC, and did the antibiotic 15.9% use the same assay gate as the expert workflow?