Ranomics
Glycosylated protein target with attached N-linked glycans and a designed Boltz-2 binder docked at a modified-residue interface
Binder design

BoltzGen glycan-aware binder design

BoltzGen is Boltz-2 binder design with native support for glycans, post-translational modifications, and non-canonical amino acid residues. The features other de novo binder design tools quietly strip out before generation. Every candidate is refolded end-to-end on the same Boltz-2 model used for design.

Hosted self-serve at tools.ranomics.com. No CCD-stripping, no PTM reversion, no manual preprocessing.

For targets where chemistry matters: antibody Fc with N-glycans, viral spike glycoproteins, phosphorylated kinase regulators, mucins.

How it works

From glycosylated target to refolded binders in one pass

01

Define the target

Upload a PDB or mmCIF with glycans, PTMs, or non-canonical residues intact. CCD codes (NAG, MAN, SEP, TPO, MLY, etc.) are passed straight to Boltz-2. No manual stripping required.

02

BoltzGen generation

Boltz-2 samples binder candidates conditioned on the full chemical environment of the target. Modified residues and attached glycans are visible to the model during design, not erased.

03

End-to-end refold

Each generated complex is refolded by the same Boltz-2 model and scored on ipTM, pLDDT, and PAE. No swap to a separate validator that may not understand the modifications.

04

Ranked candidates

Top designs returned with confidence metrics, designed sequence, and a PDB of the predicted complex. Ready for ProteinMPNN refinement, yeast display validation, or downstream wet-lab handoff.

Methodology

Boltz-2 as both generator and validator

BoltzGen is built on Boltz-2 (Passaro et al., 2025), the open-source biomolecular foundation model that follows Boltz-1 (Wohlwend et al., bioRxiv 2024, "Democratizing Biomolecular Interaction Modeling"). Five components define what BoltzGen does that other generators do not.

Foundation

Boltz-2 architecture

A pair-representation transformer trained on PDB complex structures with affinity heads added. AlphaFold3-class accuracy, fully open weights, and joint modeling of structure and binding. The same model writes the design and grades the fold.

Native

Glycans as first-class entities

N-linked and O-linked glycans (NAG, MAN, BMA, FUC, SIA, GAL and the full CCD library) attached to the target are visible to Boltz-2 during generation. The model designs around them rather than pretending they are not there.

Native

PTM-aware design

Phosphorylation (SEP/TPO/PTR), acetylation, methylation, hydroxylation, and any other CCD-coded post-translational modification on the target chain is passed through to the model. Binders can be designed against the modified state directly.

Native

Non-canonical residue support

Selenomethionine, pyrrolysine, hydroxyproline, and other modified amino acids in the target are handled natively. Useful for engineered or synthetic protein targets that other binder design tools cannot ingest without manual mutation back to canonical residues.

Validation

End-to-end refold scoring

Every generated complex is refolded by Boltz-2 itself and scored on ipTM (interface predicted TM-score), pLDDT (per-residue confidence), and PAE (predicted aligned error). No hand-off to a separate validator that may have been trained on different chemistry.

Why this matters

Glycan-aware design is not optional for half of pharma targets

When chemistry shapes the surface

Antibody Fc engagement, viral spike glycoprotein neutralization, mucin-domain targeting, and phospho-regulated kinase scaffolds all have non-amino-acid features that dominate the binding interface. Strip them and you are designing against a protein that does not exist in cells.

What other generators ignore

RFdiffusion and BindCraft both require glycans stripped and PTMs reverted to the parent residue before generation. The resulting binders may clash with the real glycoform once expressed. BoltzGen sees the target chemistry the cell sees.

One model, design and validate

Most binder design pipelines pair a generator (RFdiffusion) with a separate validator (AF2 multimer). BoltzGen uses Boltz-2 for both, so the confidence score on a candidate reflects the same model and chemistry that produced it. No train-test mismatch on modifications.

When to use BoltzGen

The right tool when the target is more than a peptide chain

For a clean cytoplasmic target with no modifications, RFdiffusion or BindCraft are the workhorses. BoltzGen earns its place when the target carries glycans, post-translational modifications, or non-canonical residues that matter for the binding interface.

Run BoltzGen first when the modifications are at or near the candidate epitope. Combine BoltzGen with RFdiffusion outputs when you want maximum diversity and the modifications are peripheral.

Antibody Fc engineering where N297 glycosylation drives FcγR engagement

Viral spike glycoproteins (HIV Env, SARS-CoV-2 S, influenza HA) with dense N-glycan shielding

Phosphorylated kinase regulatory loops where pSer/pThr is the recognition motif

Mucin-domain targets where O-linked glycans define the surface

Engineered protein targets containing selenomethionine, hydroxyproline, or other non-canonical residues

Acetylated histone tails and other PTM-marked epigenetic readers

Any target where the published co-crystal structure includes ligands or modifications you need to design around

Design against the target your cells actually express

Create a free tools.ranomics.com account and run BoltzGen on your glycosylated, phosphorylated, or non-canonical target. Refolded results, ranked and ready for handoff.