Protein Sequence Design PipelineThe protein sequence design was conducted using the Foundry toolset (incorporating RFdiffusion, ProteinMPNN, and RoseTaffaFold). The workflow followed these iterative steps:Phase 1: Initial Design and ScreeningInitial Generation: Protein sequences were generated, and a preliminary screening was performed based on pAE (Predicted Alignment Error) and pLDDT (Predicted Local Distance Difference Test) metrics.Backbone-Constrained Relax: The filtered structures underwent a single round of Rosetta Relax with backbone constraints.Stability Filtering: Structures that achieved a binding energy change of $\Delta \Delta G < -50$ passed this stage.Unconstrained Refinement: These successful candidates were subjected to 10 iterations of Rosetta FastRelax without constraints.Phase 2: Iterative OptimizationTemplate Selection: Complex files with the highest energy scores (lowest total energy) were selected to serve as the backbones for the next design cycle.Redesign: The design process was repeated, followed by a second round of pAE and pLDDT screening.Final Validation: After another round of FastRelax, candidates were filtered by their final $\Delta \Delta G$ values to identify the most stable designs.