How to Use RFantibody

RFantibody is a de novo binder-design pipeline for antibodies and nanobodies that uses structure generation rather than library screening as its starting point. The method combines fine-tuned diffusion backbone design with inverse folding and structural self-evaluation, making it useful when the scientific question is how to build a new binder against a known protein surface rather than how to rank an existing sequence panel.

On Neurosnap, researchers provide an Input Framework, an Input Target, and optionally Hotspots that define the epitope they want to engage. Mode switches between antibody and nanobody campaigns, while Design Loops allows loop-specific redesign when the goal is focused CDR engineering instead of a full variable-domain search.

RFantibody is a multi-stage design workflow. A fine-tuned RFdiffusion model first proposes backbone-level antibody or nanobody geometries in the context of the target surface, ProteinMPNN then sequences those candidate backbones, and a specialized RoseTTAFold2 oracle is used to check whether the designed binders remain structurally self-consistent and target focused. That combination is what makes the method more than a generic sequence generator.

On Neurosnap, Number of RFdiffusion Backbone Designs controls how many independent structural hypotheses are explored and Number of ProteinMPNN Designs determines how many sequences are sampled for each backbone. Together with Hotspots and Design Loops, those settings define whether the run behaves like a narrow epitope-focused maturation campaign or a broader de novo search around a target patch.

Researchers should treat RFantibody output as a ranked design set. The most useful comparisons are usually between candidate interface geometry, loop placement, structural confidence, and sequence diversity before moving a short list into folding validation, developability screening, or wet-lab expression.

• Study-fit structural inputs: RFantibody starts from both a target structure and a framework template, which is how many real antibody-engineering campaigns are framed.
• Multi-stage design logic: RFdiffusion, ProteinMPNN, and a structural oracle contribute distinct checks on backbone generation, sequence choice, and plausibility.
• CDR-level control: Mode, Hotspots, and Design Loops let researchers focus the search on the exact binder format and interface region they care about.
• Campaign-ready triage: Neurosnap presents RFantibody as a candidate-generation workflow rather than a single black-box answer, which is more useful for experimental decision-making.