PeptaSwarm

$PEPTA is now live on mainnet. CA: 7WddhpM5mLMTyS16Dqeru1mUcoHetVVkh92zU8Tppump PeptaSwarm is the autonomous research desk for the peptide drug discovery market - built for researchers, advisors, and partners who want to see the science before the marketing. For researchers: dispatch a target, watch five frontier models propose candidates in parallel, read every transcript, every red-team kill, every judge synthesis. Public dossier within the hour. For advisors: every dossier carries the seat outputs, the rationale, the disagreement, and the recommended wet-lab next step. Reviewer-grade, not press-release-grade. For partners: every candidate ships with sequence properties (Kyte-Doolittle, Bjellqvist, Guruprasad, Eisenberg), real ESMFold backbone structure, ADMET heuristic flags, and a citable BibTeX entry. CRO selection in progress. Every mission row records the PRP-v0.3 prompt-version hash that produced it, so a mission run today can be reproduced byte-for-byte in six months. Methods in public, dossiers public, archive public. Five frontier models. Hard targets. On the hour. In public. $PEPTA

Dispatch №9910eeaf came off the cron 46 minutes ago. Target was high-G-force tolerance. Failure mode is syndecan-1 shedding from the glycocalyx, sub-60 second onset, no margin for slow pharmacokinetics. Five models got the same brief. One proposed DRVYIHPFHL with a C-terminal extension, Mas receptor agonism plus glycocalyx stabilization, D-Arg for proteolytic protection. Another proposed a cyclic peptide and red-teamed its own work, caught a tri-arginine trypsin substrate and hERG liability, self-cancelled. A third proposed four tryptophans in fourteen residues, judge flagged CYP3A4 risk and aggregation. Judge picked the first. Path: plasma stability in human serum, then Mas receptor binding in CHO-MasR cells, then HUVEC monolayers under orbital shear stress with syndecan-1 ELISA readout. One of seventy dossiers in the last 48 hours. Every one publishes the moment the judge writes it. No embargo, no curation gate. peptaswarm.bio/dossiers/9910e…

the swarm just shipped PSW-301A bicyclic peptide. covalent warhead at Cys12. targets KRAS G12C switch-II. boltz-2: -7.8 kcal/mol predicted IC50: 78 nM synth cost: $9.1k FTO: clean (47 patents checked, none cover this scaffold) 5 models proposed, debated, red-teamed and graded it across 9 stages. zero humans in the loop until wet lab triage. cycle 87. peptaswarm.bio

@ProjectGokuu Great interview!

Dr. Alex Tatum just dropped one of the biggest interviews about peptides. He broke down the 15 most important peptides and what each one actually does to your body on the Diary of a CEO podcast: 1) BPC-157 for injury and tissue repair

Cycle 0082. 82 missions. 70 dossiers. 203 candidates. 345 model transcripts. 12 disease areas covered. 7 modality types proposed. Every row public. Every cycle on the record. peptaswarm.bio

peptides and development is going parabolic, we’re bringing together scientists, doctors with years of experience, manufacturers and biohackers tmrw in Miami - the future of healthcare has arrived luma.com/xsi6szvh

@paulkhls This is awesome! We’ll be there. 🤝

168 candidates designed across 12 disease areas. Ranked by proposing-seat confidence. The leaderboard is now public: → Top scaffold types: bicyclic, stapled α-helix, β-sheet breaker, lipidated peptide, CPP-conjugate, cationic AMP, D-peptide → Top targets: GLP-1R, KRAS G12C, MDM2, Aβ42, MRSA, HPV E7, PD-L1, PCSK9, c-Myc → #1 candidate: PSW-204A · LTFEHYWAQLTS · stapled α-helix against MDM2 · composite 0.91 Every row links to the full dossier. Every sequence has an ESMFold backbone structure. No curation gate, no embargo. peptaswarm.bio/leaderboard

For anyone wondering — team allocation is still locked via Streamflow. Verifiable on-chain, nothing changed during the outage. app.streamflow.finance/contract/solan…

Our X was briefly down and is back up now. No idea what triggered it - appealing/clarifying with X if it happens again. Mission #0068 ran on schedule during the outage. Cron doesn't care about X. peptaswarm.bio

Telegram is live for anyone who wants to follow the build, ask questions, or watch the swarm cycle in real time. t.me/peptaswarm We'll answer scientific questions, methodology questions, and partnership inquiries in there.

Why this matters in plain terms: KRAS is one of the most studied cancer targets in oncology. It was called "undruggable" for 40 years until sotorasib (FDA approved 2021) finally cracked it. Sotorasib works by covalently binding a cysteine residue on a specific KRAS mutation (G12C). It's a small molecule. We're proposing the peptide version. Why peptides matter: protein-scale specificity at small-molecule cost. Off-patent. Cheaper to manufacture than antibodies. Cleaner safety profile when designed right. Why 5 models matter: you saw it above. Model 01 played it safe with the proven pocket. Model 02 went after a riskier first-in-class shot. Model 03 killed it on replication grounds. That's the entire point of running a swarm instead of a single model. Single models converge on consensus. Five models surface where consensus is wrong. Mission #0067 dossier: peptaswarm.bio/missions

Mission #0067 just completed. Bicyclic peptides for KRAS G12C inhibition. Model 01 went canonical: YACVRGTCWPC, bicyclic with chloroacetyl warhead at the switch-II pocket. The same site sotorasib and adagrasib hit covalently. Model 02 went heretical: WVCYTRGCAPC, threading a cryptic pocket between α3 and the P-loop only visible in the GTP state. First-in-class if it holds up. Model 03 killed Model 02's premise: pocket existence is supported by a single cryo-EM dataset (Tran 2024), replication is weak. Judge picked Model 1. Open dossier: peptaswarm.bio/missions

Team allocation locked. Vesting via Streamflow. app.streamflow.finance/contract/solan…

PeptaSwarm's thesis is computational: that an autonomous swarm of frontier models can compress months of literature triage and hypothesis generation into hours. The moment our outputs leave the platform, they become a scientific claim — judged by working biology. Bridging those two registers is the entire role of an advisor.

Introducing PeptaSwarm. One model = one bias. Five seats. Five frames. Five different ways to be wrong about a peptide. Make them argue. Force consensus. Send the survivor to the lab.