A first look at onepot CORE for rapid hantavirus antiviral discovery

From a newly shared patient-derived Andes virus genome to synthesizable molecules for experimental testing

This morning, the sequence appeared

This morning, a patient-derived Andes virus genome was posted publicly. Within hours, we began using that sequence to launch a synthesis-aware antiviral screen from onepot CORE.

That is the point of this live worklog. onepot is not a virology company, and we do not usually originate the biology. But when a biological starting point exists, the path from computational prioritization to real, purified, QC'd compounds should be much shorter. Most computational outbreak-response work stops at predictions. We want to help it stop at compounds.

The sequence became available today, May 8, 2026, through a public post on Virological.org. We are grateful to the Swiss National Reference Center for Emerging Viral Infections, Geneva University Hospitals, the Institute of Medical Virology at the University of Zurich, and the authors who shared the consensus sequence openly.

This is not a claim that we have discovered a treatment. It is an example of how public sequence data can be connected to a synthesis workflow quickly, while the biology is still being interpreted. If you have relevant biology, targets, or assays where rapid compound synthesis could help, reach out at hello@onepot.ai .

Hantaviruses can cause severe disease, and Andes virus is especially important because it is one of the few hantaviruses associated with person-to-person transmission. Outbreak-linked sequences give researchers a concrete starting point. The hard part is not only identifying targets computationally, but turning those targets into chemistry that someone can actually test.

What we did

A viral genome is not a drug target by itself. But it gives us the amino acid sequences of viral proteins. Those sequences can be compared, modeled, and used to prioritize regions where small molecules might plausibly bind.

At a high level, the workflow is straightforward: extract viral protein sequences, identify plausible antiviral targets, prepare a synthesis-aware compound pool from onepot CORE, run structure-aware affinity prediction, and prioritize molecules that combine computational signal with practical synthesis feasibility.

We treat predicted affinity values as enrichment signals, not answers. Molecules become interesting only when computational signal, chemical diversity, and a credible synthetic route point in the same direction.

What we targeted

The viral L protein is the first live screening target because polymerase inhibition is a known antiviral strategy across many RNA viruses, and the L protein contains metal-dependent enzymatic machinery. For hantaviruses, the challenge is structural uncertainty and assay validation. This makes the L protein a good place to generate hypotheses, but a poor place to overstate confidence.

Viral entry machinery is also compelling, but structurally dynamic. Modeled glycoprotein regions may be useful for proposing experiments, especially with the right assay partner, but they are lower-confidence small-molecule targets. Nucleocapsid proteins may expose functional protein-RNA or oligomerization interfaces, though these are less straightforward and belong in an exploratory bucket.

Why onepot CORE helps

The point of screening the onepot CORE compound space is not only scale. The point is that the molecules are connected to the chemistry we can run.

Virtual screening often ends with a list of molecules that are hard to buy, hard to make, or disconnected from the lab. onepot CORE is designed differently: every molecule is tied back to a synthesis plan. That matters in outbreak-relevant science because the useful output is not a docking score. The useful output is a short list of molecules that someone can actually test.

Hit molecules can be prioritized by predicted binding, synthesis feasibility, delivery time, price, and chemical diversity. The output is closer to an experimental queue than a static virtual library.

First screen results

We ran a six-iteration structure-aware affinity screen from onepot CORE. The technical setup used Boltz-2 affinity prediction against the Andes virus L-protein sequence with an Mn cofactor, a 2.7M-compound CORE v1.1 subset, RDKit filters for basic drug-like properties and liability flags, and a random initial batch followed by active learning on Morgan fingerprints.

The campaign screened 4,772 compounds, corresponding to 3,874 unique structures, in roughly 85 minutes of wall time using 150-200 GPUs in parallel. After filtering a 2.7M-compound subset down to 688,677 molecules, the active-learning loop identified 219 unique compounds with predicted log10(IC50) below -0.5. Those are computational affinity signals, not antiviral activity.

The best predicted compound appeared in the first active-learning round at log10(IC50) = -1.481, about 33 nM on the model's scale. The most useful engineering result, though, was the final greedy pass over the full pool: it found 97 of the 219 predicted strong binders, more than the four UCB rounds combined. That is the practical lesson for synthesis triage: explore enough to train a useful surrogate, then exploit broadly enough to find makeable follow-up sets.

What needs validation

The next step is experimental. For each selected molecule or cluster, the useful payload is the predicted target, docking rationale, CORE reaction class, estimated synthesis feasibility, expected turnaround, and the assay that would actually test the hypothesis.

For qualified collaborators, we can contribute the synthesis side of the experimental plan: prioritized molecules under appropriate terms, synthesis routes, physical compounds for testing, and follow-up analogs from the CORE compound space, while assay design and biological interpretation stay with the teams closest to the biology.

Who should collaborate

We are not looking for press around unvalidated computational hits. We are not trying to replace the teams closest to the biology. We are looking for collaborators with targets, assays, or urgent antiviral questions where fast access to synthesized compounds could quickly create real data.

We would especially like to talk with groups working on hantavirus or Andes virus biology, structural virology, polymerase assays, minigenome or replicon systems, pseudovirus entry assays, BSL-appropriate antiviral screening, and outbreak-response drug discovery. If you already have biological context and need a fast path to synthesized compounds, please reach out at hello@onepot.ai .

Why this matters

The broader lesson is not specific to this sequence. Modern biology can share genomes quickly. Computational tools can model proteins quickly. But chemistry often remains slow, fragmented, and disconnected from execution. We at onepot want to close the synthesis gap: from a prioritized molecule to a real compound fast enough that experimental collaborators can actually use it.

This is an early computational screen, not a therapeutic result. But it is exactly the kind of workflow we think should exist: public sequence data, rapid modeling, synthesis-aware molecule selection, and fast experimental follow-up — together enabling faster DMTA loops and helping deliver medicines faster to everyone who needs them.