How to Use BAGEL Protein Design

BAGEL is a programmable protein design framework that treats protein engineering as exploration over an energy landscape rather than as one fixed redesign trajectory. On Neurosnap, the BAGEL Protein Design service exposes three user-driven workflows: Mini-Enzyme mode for structure-based compaction around a preserved catalytic neighborhood, Mimic-Enzyme mode for sequence optimization with immutable positions, and Binder mode for designing a mutable binder against a target sequence and hotspot definition.

This broader setup is useful when researchers need direct control over what is protected and what is allowed to change. Instead of relying on preset templates, users supply the structure or sequence inputs, identify critical residues, hotspot regions, or immutable positions, and then tune the optimization schedule to match the experimental question.

The BAGEL paper frames protein engineering as sampling over an objective built from protein-language-model embeddings together with explicit penalties that shape the search, including a size-related term for compaction. In practice, BAGEL performs Monte Carlo moves through sequence space and accepts or rejects substitutions, insertions, and deletions according to the objective and the selected temperature. That makes the method better suited to exploratory redesign than tools that only score predefined mutants or only permit conservative substitutions.

On Neurosnap, Mode determines the scientific setup. Mini-Enzyme mode starts from an Input Structure and lets users define Optimization Windows, Critical Residues, and a Conservation Buffer, which is useful when only part of a structure should be redesigned and catalytic residues must remain protected in chain-qualified numbering. Mimic-Enzyme mode starts from an Input Sequence and preserves explicitly declared Immutable Residues, making it more appropriate for sequence-level enzyme redesign without requiring a structure input. Binder mode uses a Target Sequence, Hotspot Residues, and Binder Length to search for a mutable binder sequence focused on a user-specified target patch.

Shared controls such as Number of Steps, Temperature, Mutations Per Step, and Log Interval determine how broadly BAGEL explores the landscape. Mode-specific controls such as substitution, insertion, and deletion probabilities or Chemical Potential Weight further shape whether the search favors compactness, conservative edits, or more exploratory sequence changes. Researchers should review the resulting sequences, energies, and optimization logs as a ranked design panel for downstream filtering rather than as a single definitive answer.

• Three distinct BAGEL workflows: Mini-Enzyme, Mimic-Enzyme, and Binder modes cover structure-based compaction, sequence-only enzyme redesign, and target-guided binder design in one service.
• Explicit residue-level control: Users can define optimization windows, critical residues, immutable positions, and hotspot regions instead of being constrained to fixed presets.
• Programmable Monte Carlo search: Temperature, step count, mutation frequency, move probabilities, and size-penalty settings make the exploration strategy adjustable to the biological question.
• Research-ready outputs: Exported energies, sequences, and optimization logs support comparative triage, iterative reruns, and downstream structural or experimental follow-up.