The AI Co-Scientist
for drug development.
Helixir generates novel hypotheses, runs them through a scientific harness of ML models and scientific tools, and returns novel candidates. We do everything up to the wet lab, and every claim traces back to its source.
Watch it generate
something new.
Pick a program and Helixir generates: new hypotheses, new candidates, new trial designs. Runs are red-teamed, reviewed and ranked, and come back as reviewable artifacts where every claim traces to its source. Methods differ by program.
helixir.deepbioscientific.com/co-scientist/protein-designillustrative run| # | Candidate | ΔG | Kd (pH 7.4) | Release | Evidence | Prov. |
|---|---|---|---|---|---|---|
| 1 | PD1L1-452 | −11.3 | 0.37 nM | 77× | 37 | |
| 2 | PD1L1-309 | −10.8 | 0.52 nM | 46× | 29 | |
| 3 | PD1L1-118 | −10.1 | 0.91 nM | 34× | 22 |
| # | Site | Conservation | Druggability | Prov. |
|---|---|---|---|---|
| 1 | Site α-4 | 98% | 0.86 | 19 |
| 2 | Loop β-7 | 94% | 0.71 | 14 |
| 3 | Pocket C | 91% | 0.66 | 12 |
| # | Structure | TM-score | Identity | Prov. |
|---|---|---|---|---|
| 1 | PDB 6XY-A | 0.91 | 42% | 8 |
| 2 | PDB 5AB-C | 0.88 | 38% | 6 |
| 3 | PDB 7KL-M | 0.84 | 35% | 5 |
Helixir generates Biology
Most AI for biology retrieves and summarises what is already known. Helixir generates novel hypotheses and de novo designs, produced by a scientific harness (first of its kind) that runs ML models and scientific tools together, and backed by evidence you can audit.
A generative program, not a literature search.
We design the molecule, not just the answer.
Give Helixir a target and it invents binders that do not exist yet, folds them, scores them on predicted affinity, stability and developability, and hands back the candidate as a structure you can turn. Not a literature summary. A molecule, with its evidence attached.
- Novel sequences generated, then folded and ranked
- Predicted properties shown with their limitations
- Binding pocket and interface open to inspection
Structure viewerliveNovel, and still grounded.
Anything can generate a plausible molecule. Ours is driven by a harness that runs ML models and scientific tools over a graph of ninety-plus sources: genes, targets, assays, trials, structures. The graph is where a new idea is built from, and what it is checked against.
- Ninety-plus integrated sources, evidence-typed
- Negative and conflicting findings kept, not discarded
- ML models and scientific tools driven together, not a single model
Knowledge GraphliveFrom open question to testable candidate.
Every run returns an artifact a scientist can argue with: the claim, the evidence for and against it, what it does not establish, and the next experiment worth running. We take a programme all the way to the wet lab.
Claim → evidence → next experimentExample trace · illustrativeRun your own programs
Early access for researchers and biotech teams. Bring a target or an open question and get reviewable, evidence-backed artifacts back.
See the platform →Partner on a pipeline
Custom design programs with pharma and biotech partners: higher-value engagements with milestone-based upside.
Talk to us →What we have shipped and published.
Early access to Helixir opens to research teams
Research and biotech teams can apply for early access to the Co-Scientist and run hypothesis discovery, de novo design and literature review on their own targets.
Read →MIRAGE: a benchmark for protein-family generalization in affinity prediction
Our preprint treats historical public protein-family support as an explicit experimental variable, for affinity and for pose, and reports what happens to frontier co-folders on families they have not seen.
Read →OIL: an open codon-resolved nucleotide framework, presented at MoML 2026
The Omics Integrated Library assembles 2.35 trillion nucleotides across fifteen sources, codon-resolved and with provenance on every record. Scripts and the automated procedure follow later this year.
Read →DeepBio Scientific builds Helixir, an AI co-scientist for drug development. It generates novel hypotheses and de novo designs, takes them through generation, red-teaming, peer review and ranking, and returns reviewable, evidence-backed artifacts. It covers everything up to the wet lab.
Most of them retrieve and summarise: they find what is already written and compress it. Helixir generates new hypotheses and new candidates, then red-teams what it generated and hands back the evidence and the disagreement along with the result. The test is whether the output is something nobody had written down yet, and whether you can audit how it got there.
Hypothesis discovery, de novo protein design, clinical trial design, literature review, conservation discovery, and structural homolog search, each returning an exportable artifact grounded in evidence.
Research and biotech/pharma teams adopting it for their own programs (via early access or custom design partnerships), and investors evaluating the thesis and the working product.
Explore the platform to see the co-scientist, the knowledge graph and the live instruments, or apply for early access and run your own programs.
De novo drug design, everything up to the wet lab.
Whether you run programs or back the teams that do, DeepBio Scientific is building the co-scientist that invents, and shows its work.