Responsible research
Designing molecules with computational assistance raises obligations that go beyond doing good science. This page sets out how we work and what we will not do.
Scope of our work
We design nucleic acid binders, specifically RNA and DNA aptamers and oligonucleotides, intended as therapeutics, diagnostics and research reagents. Our programmes target human disease biology: secreted proteins, cell-surface receptors, coagulation factors and structured viral RNA elements.
We do not work on enhancing the transmissibility, virulence or host range of any pathogen. We do not design agents intended to cause harm, and we do not accept collaborations whose purpose is weapons development. Where a target has a plausible dual-use dimension, such as a viral element, the work is directed at inhibition and detection, and we assess the risk before a programme starts rather than after.
Biosecurity
Designed sequences are nucleic acid, so the relevant safeguards are those of the synthesis supply chain. We order synthesis through established providers that screen orders, and we do not attempt to circumvent screening by splitting or obfuscating sequences. Our candidates are short, non-coding, non-replicating oligonucleotides, which places them well outside the categories of concern for synthetic genomics, but we treat screening as a floor rather than a formality.
Where a design campaign touches a pathogen-derived target, we work with the relevant institutional biosafety oversight at our collaborating laboratories, and experimental work proceeds under their approvals and containment requirements, not ours alone.
Human oversight of automated design
Our platform uses agentic systems to reason about targets, generate candidates and analyse results. Those systems prepare and recommend. They do not decide.
- A scientist approves the constraint specification before any generation begins.
- A scientist approves the synthesis shortlist. Nothing is ordered automatically.
- Every recommendation carries its reasoning, the constraints it satisfies and the tool scores behind it, so that a review can engage with a specific step rather than a conclusion.
- Where tools disagree, the disagreement is surfaced rather than averaged away.
We consider the audit trail a safety feature, not a compliance artifact. An automated system whose outputs cannot be interrogated should not be trusted to prioritize experiments.
Data integrity
The loop only works if the measurements are honest. We record assay context with every result, including instrument, buffer, folding protocol, immobilization density and replicate structure, because a number without its method cannot be reused.
Negative results are retained and reported. We do not discard datasets that contradict a design hypothesis, and we do not present a selected subset of rounds as representative of a campaign. Where we identify an artifact after the fact, we correct the affected analysis and say that we have done so.
Research ethics
Our current work is computational and in vitro. Any future work involving animals would proceed only under institutional animal care and use approval at the performing institution, with explicit attention to reduction and refinement. Any future work involving human subjects or human-derived samples would proceed only under institutional review board approval and informed consent, in accordance with Korean law and the Declaration of Helsinki.
Scientific communication
We describe our programmes as preclinical and investigational, because they are. We do not imply regulatory approval, clinical validation or therapeutic benefit that has not been demonstrated, and we aim to publish methods including the cases where our approach performed poorly. A field that only hears about successes converges more slowly on what actually works.
Reporting a concern
If you believe our work raises a safety, security or ethical concern, write to hello@sanghyun.me with "Responsible research" in the subject line. We will acknowledge within five business days. Concerns can be raised anonymously, and we will not seek to identify a person who raises one in good faith.
Changes
We will update this page as our programmes and obligations change, and we will revise the effective date above when we do.