Introducing Origin-1:
a platform for de novo antibody design against "zero-prior" epitopes
Origin-1 is a generative AI platform that designs full-length monoclonal antibodies (mAbs) against "zero-prior" epitopes (target sites with no reported complex structures of antibody or protein binders). Origin-1 delivers specific, developable binders in fewer than one hundred designs per target.
<100 Designs per Target
5+ Orthogonal Assays
Atomic Accuracy by Cryo-EM
~100nM Functional Potency
The Scientific Challenge
Many de novo antibody design models have succeeded in delivering binders against solved epitopes (binding sites that have already been structurally characterized with a known protein binder). While useful, these approaches do not prioritize a critical problem in therapeutic discovery: designing antibodies for targets and epitopes lacking solved complex structures with protein binders.
Origin-1 is an AI platform developed and validated to address this challenge.
Targeting "Zero-Prior" Epitopes
To demonstrate Origin-1's generalizability, we selected targets and epitopes with no antibody-antigen or protein-protein complexes in the Protein Data Bank (PDB). In addition, all homologous complexes were excluded from training, validation, and test sets.
We define these putative epitopes as "Zero-Prior" Epitopes: binding sites on target antigens for which prior structural definition of proteins, including antibodies, in complex with these targets is lacking in public databases. These epitopes were selected based on surface geometry, chemistry, and biophysical properties along with literature review.
The Model Breakthrough
Platform Architecture
Origin-1, a design-and-score platform
- AbsciGen (The Generative Engine): Generates antibody structures and sequences
- Models:
- AbsciDiff: An all-atom diffusion model that conditions on unbound antigen structure, antibody frameworks, and user-defined epitopes to generate antibody-antigen complexes and design novel binding modes
- IgDesign2: A structure-conditioned sequence model that integrates geometric encodings with a paired antibody language model to design all six complementarity determining regions (CDRs), modeling coupling of the heavy and light chains while improving humanness and developability
- Models:
- AbsciBind (The Scoring/Filtering Engine): Scores and filters antibody candidates prior to experimental characterization
A scalable AI x Bio engine. Origin-1 was developed using multiple years of proprietary data generation. Our "Lab-in-the-Loop" platform continuously feeds wet-lab validation data back into the models, capturing the fundamental binding mechanics and rules often absent from public datasets.
Millions of Data Points
Years of rapid validation loops
Key Results
Validated on "Zero-Prior" Epitopes across Four Targets
Origin-1 was tested against "zero-prior" human protein epitopes across ten targets, and was successful for four: COL6A3, AZGP1, CHI3L2, and IL36RA.
Successful Targets and Relevance:
- COL6A3 – Extracellular matrix
- AZGP1 – Lipid metabolism
- CHI3L2 – Inflammation
- IL36RA – Immune signaling
Atomic Accuracy
We solved the structures of our top binders against COL6A3 and AZGP1 using cryogenic electron microscopy (Cryo-EM). The experimental structures matched the design models closely, confirming Origin-1's ability to generate epitope-specific, atomically-accurate binders.
| Target | Global RMSD (All-Atom) | HCDR3 RMSD (All-Atom) | DockQ | Resolution |
|---|---|---|---|---|
| COL6A3 | 2.56Å | 0.66Å | 0.83 | 3.0Å |
| AZGP1 | 1.79Å | 1.41Å | 0.73 | 3.1Å |
Functional Antagonism
For IL36RA, a zero-shot binder was optimized using AI-guided affinity maturation, achieving up to 68-fold affinity gain, and functionally antagonizing with ~100nM EC50.
Demonstrated Efficiency
In contrast to traditional screening methods, Origin-1 generated potential lead candidates by screening fewer than one hundred designs per target.
To our knowledge, this is the first demonstration of de novo design of full-length, monoclonal antibodies (mAbs) against "zero-prior" epitopes with atomically accurate complex structures and functional activity.
Therapeutic-Grade Validation
Origin-1 adheres to pharmaceutical lead identification standards. We employ stringent multi-assay validation criteria to ensure confidence in therapeutic translation. A design is strictly classified as a success only if it meets all of the following requirements:
Binding & Specificity
- Hits identified with Surface Plasmon Resonance (SPR) using both automated criteria and expert review
- Demonstrates specific binding to the intended target with no detected binding to off-target proteins
- Demonstrates binding in orthogonal Biolayer Interferometry (BLI) assay in both mAb and Fab format
- Demonstrates complex formation measured by analytical Size Exclusion Chromatography (SEC), confirming antigen engagement in solution
Developability Screen
- Passes initial filters for aggregation, polyreactivity, hydrophobicity, and stability
With this standard and <100 designs per target entering the lab, Origin-1 produces specific, developable antibodies against multiple "zero-prior" epitopes. The emphasis is on antibody quality and reliability.
Accelerating Therapeutic Discovery via Transparency
Absci is committed to full methodological, architectural, and data transparency, ensuring our findings are both rigorous and reproducible. By providing this detailed technical foundation, we empower the broader research community to accelerate the development of biologics against hard-to-drug targets.
