Design antibody CDR regions using equivariant diffusion models for de novo antibody engineering Learn more
Design antibody CDR regions using equivariant diffusion models for de novo antibody engineering Learn more
Design antibody CDR regions using equivariant diffusion models for de novo antibody engineering Learn more
Exploratory antibody CDR co-design for antibody-antigen complexes using Proteo-R1 reasoning and raw diffusion. The standard online workflow does not include the framework structure-inpainting assets required for the published-quality target.
Structure-based de novo antibody and nanobody design pipeline combining antibody-tuned RFdiffusion, ProteinMPNN sequence design, and antibody-tuned RoseTTAFold2 filtering.
Generate protein structures and scaffolds with Genie 3, an all-atom SE(3)-equivariant diffusion model. Genie 3 supports unconditional protein generation, motif scaffolding, and hotspot-targeted binder design.
All-atom generative AI for designing protein binders. Specify target binding sites and generate diverse binding proteins with fine-grained control over interaction parameters.
ProGen2 is Salesforce Research's protein language model suite for prompt-based de novo protein sequence generation and bidirectional sequence likelihood scoring.
Antibody humanization and humanness evaluation platform from Merck. Sapiens mode uses deep learning trained on the Observed Antibody Space (OAS) to humanize antibody sequences, while OASis mode evaluates humanness using 9-mer peptide search against human antibody databases.
BoltzGen uses generative diffusion models to design protein, peptide, nanobody, and Fab binders against protein and small-molecule targets.
ProFam-1 is a protein family language model for family-conditioned sequence generation. Provide a protein family in FASTA, A2M, or A3M format and generate new sequences with model likelihood scores for downstream ranking and screening.
EvoDiff is a diffusion-based protein sequence generation framework from Microsoft Research. ProteinIQ currently runs the EvoDiff-Seq OA_DM_38M model for unconditional protein generation, motif scaffolding, and user-sequence inpainting.
PocketFlow is a structure-based molecular generative model that designs novel drug-like molecules within protein binding pockets. It uses autoregressive flow modeling with chemical knowledge to generate 100% chemically valid, highly drug-like compounds.
Exploratory antibody CDR co-design for antibody-antigen complexes using Proteo-R1 reasoning and raw diffusion. The standard online workflow does not include the framework structure-inpainting assets required for the published-quality target.
Structure-based de novo antibody and nanobody design pipeline combining antibody-tuned RFdiffusion, ProteinMPNN sequence design, and antibody-tuned RoseTTAFold2 filtering.
Generate protein structures and scaffolds with Genie 3, an all-atom SE(3)-equivariant diffusion model. Genie 3 supports unconditional protein generation, motif scaffolding, and hotspot-targeted binder design.
All-atom generative AI for designing protein binders. Specify target binding sites and generate diverse binding proteins with fine-grained control over interaction parameters.
ProGen2 is Salesforce Research's protein language model suite for prompt-based de novo protein sequence generation and bidirectional sequence likelihood scoring.
Antibody humanization and humanness evaluation platform from Merck. Sapiens mode uses deep learning trained on the Observed Antibody Space (OAS) to humanize antibody sequences, while OASis mode evaluates humanness using 9-mer peptide search against human antibody databases.
BoltzGen uses generative diffusion models to design protein, peptide, nanobody, and Fab binders against protein and small-molecule targets.
ProFam-1 is a protein family language model for family-conditioned sequence generation. Provide a protein family in FASTA, A2M, or A3M format and generate new sequences with model likelihood scores for downstream ranking and screening.
EvoDiff is a diffusion-based protein sequence generation framework from Microsoft Research. ProteinIQ currently runs the EvoDiff-Seq OA_DM_38M model for unconditional protein generation, motif scaffolding, and user-sequence inpainting.
PocketFlow is a structure-based molecular generative model that designs novel drug-like molecules within protein binding pockets. It uses autoregressive flow modeling with chemical knowledge to generate 100% chemically valid, highly drug-like compounds.
Configure inputs to begin
Set options on the left, then click “Submit job”.
DiffAb is a deep generative model that designs antibody complementarity-determining regions (CDRs) using diffusion probabilistic models and equivariant neural networks. Published at NeurIPS 2022, it was the first deep learning method to generate antibodies explicitly targeting specific antigen structures, representing an early application of diffusion models to protein design.
The model jointly generates both sequences and 3D structures of CDR loops while maintaining physical constraints through equivariance—a mathematical property ensuring predictions remain valid under rotations and translations. This antigen-aware approach distinguishes DiffAb from prior methods that lacked structural context about the target.
ProteinIQ hosts DiffAb on GPU infrastructure, providing browser-based access to antibody CDR design without local installation or environment configuration.
| Input | Description |
|---|---|
Antibody structure or antibody-antigen complex | PDB file or RCSB PDB ID. DiffAb accepts paired antibodies, heavy-chain-only nanobodies, light-chain-only antibody structures, and complexes with antigen chains. Antigen-only design requires the separate HDOCK workflow and is not available here. |
Heavy chain ID | Optional antibody heavy-chain identifier. Leave blank to let DiffAb identify and Chothia-renumber antibody chains automatically. |
Light chain ID | Optional antibody light-chain identifier. Leave blank for automatic detection or when submitting a heavy-chain-only nanobody. |
| Setting | Description |
|---|---|
Design mode | Co-design single CDR samples sequence and structure for one loop. Co-design all detected CDRs samples every detected heavy- and light-chain CDR together. Optimize one CDR starts from the native loop. Fixed backbone samples sequence while preserving backbone coordinates. Structure prediction samples structure while preserving sequence. |
Target CDR | CDR loop used by the four single-CDR modes. Options: HCDR1, HCDR2, HCDR3 (default), LCDR1, LCDR2, LCDR3. Multi-CDR mode ignores this setting and redesigns all detected CDRs. |
Optimization steps | Native-start optimization length for optimization mode. The official hosted DiffAb default is 4. |
Number of samples | How many design variants to generate (1–10, default 5). The browser limit is a compute cap; the repository test configurations use 100. |
Random seed | Reproducible inference seed (default 2022, matching DiffAb’s official configurations). Enable random seed generation explicitly for a nondeterministic run. |
DiffAb returns the source project’s native result artifacts:
DiffAb does not produce a model-confidence score or scientific ranking. Sample numbers identify generation order only. Download the PDB files to analyze structures in PyMOL, ChimeraX, or another molecular viewer.
DiffAb uses a diffusion probabilistic model that learns to denoise antibody CDR structures through iterative refinement. Starting from random noise, the model progressively generates both amino acid sequences and atomic coordinates that fit the binding site.
The neural network architecture is equivariant under SE(3) transformations (rotations and translations), meaning it treats spatial geometry consistently regardless of how the complex is oriented. This ensures physically realistic antibody conformations.
During training, DiffAb learned from antibody-antigen complexes in the Protein Data Bank, capturing patterns in how CDRs bind to different epitopes. The model supports five inference modes:
Co-design generates sequences and structures simultaneously, allowing backbone flexibility to optimize binding geometry.
Fixed backbone designs only the sequence given a predetermined structure, useful for optimizing existing antibody scaffolds.
Multi-CDR co-design jointly optimizes multiple CDR loops, capturing cooperative effects between adjacent loops that single-CDR design misses.
Antibody optimization starts from the native CDR and applies a selected number of denoising steps.
Structure prediction samples CDR coordinates while preserving the submitted amino-acid sequence.
DiffAb generates multiple design candidates. Evaluate them using:
The model outputs are starting points for optimization, not final therapeutics. Most designs require experimental testing and iterative refinement.
DiffAb maintains a rigid antigen structure during design. Real antibody-antigen interfaces often involve conformational changes upon binding that the model cannot capture.
Automatic antibody-chain detection uses AbNumber and ANARCI with Chothia numbering. If automatic detection fails, provide the chain IDs exactly as they appear in the PDB file.
Generated sequences may contain developability liabilities. The model optimizes for structural fit to the antigen, not manufacturability, stability, or immunogenicity. Additional filtering and engineering are typically required before therapeutic development.