#1 Run ProBuilder to generate scaffolds, The candidate scaffolds were first screened based on the Rosetta energy score, selecting those with a score_per_res below 0.3 Rosetta Energy Units (REU). These scaffold sequences were then subjected to the AlphaFold2 structure prediction pipeline to generate their respective predicted models. The final set of scaffolds was chosen according to a stringent criterion: the backbone root-mean-square deviation (r.m.s.d) between the design model and the predicted structure had to be less than 0.6 Å across all five AlphaFold2 models. An example scaffold was selected in "1-input_scaffold" folder. The design procedures for light-regulated homo-oligomers can be generated following the steps below. Ensure that you have the necessary ROSETTA package and related environment.

#2 Run command in '2-symmetry_scaffold' folder to generate relative AzoF interaction field for C4. Other symmetrys including C2, C3, C5 can be generated in a similar way.

#3 Run command in '3-install_AzoF' folder to install AzoF.

#4 Run 'cluster_commands.list' commands in '4-cluster' folder to.....,the parameter 'rmsd_dist' can be adjust to control the number of exported models. Then run 'extract-keys-commands.list' to extract models for FastDesigin.

#5 Run command in 'designalbe_res_command.list'in '5-fastdesign' folder to generate designalbe residues of each model for FastDesigin. Then run command in 'extract_chainA_command.list' to extract chain-A in complex. Finally, run commands in 'fastdesign_commands.list' to generate the FastDesigin output models.

#6 Run command in '6-fastdesign-score' folder to scoring the FastDesigin output models according to AzoF state. Models were filted base on scores, for examples: 'AzoF_sasa < 20; AzoF_contact > 100; ddg_with_AzoF-ddg_without_AzoF<-5.0'. The filted models were further put into ProteinMPNN design methond.

#7 Run command 'designalbe_res_command.list' in '7-mpnn' folder to generate designalbe residues of each model for FastDesigin and run command 'mpnn_command.list' to generate the ProteinMPNN model outputs.

#8 Run commands in '8-mpnn-score' folder to score the model exported from ProteinMPNN mainly based on the AzoF state, complex interaction situation and the interface hydrophobicity of these comples. Models were filted base on scores, for examples: 'score_per_res < 2.5; interface_hbonds >=3; AzoF_ddg < 7.5; AzoF_contact > 140; AzoF_neighbors > 1.5; sap_score < 40; AzoF_sasa < 15; contact_chA <1000' The filted models were further put into Alphafold2 structure prediction procedures.

#9 Run 'alphafold-commands.list' in '9-alphafold' folder to predict structures using five models. To reduce computation time, we can first calculate one model and filter it based on r.m.s.d before calculating the other four models. The final models for further selection was filter based on the r.m.s.d, for example, less than '0.9' of 5 predicted structural models.