SugarSwitch🪄

SugarSwitch is a pipeline for protein N-glycosylation modification.

📄 Paper: This is the title for the article
🌐 Webserver: This is the name for the SugarSwitch webserver

---

Quick start

Installation

1. Clone the repository

   git clone https://github.com/EricXY0706/sugarswitch.git
   cd sugarswitch

2. Run the automated setup procedure

   chmod +x setup.sh
   ./setup.sh

⚠️ Notice: Foldseek binary need to be downloaded from here and place it in the ./SaProt/foldseek folder.

✅ This procedure will automatically:

• Create and initiate sugarswitch conda environment with Python 3.10
• Install all the required dependencies
• Download SaProt and SPIRED weights
• Download and install PyRosetta and DSSP

Usage

1. Update the configrations in config.py

   • Must-ToDos:
     ➡️ Manually specify the protein chain ID to be modified with protein_chain_id and the important sites which you think the modification pipeline should ignore with functional_hotspots in user_configs

     # An example
     
     user_configs = {
      "protein_chain_id": "A", # Specify the chain ID to be modified and glycosylated.
      "functional_hotspots": [1,2,3], # e.g. [1,2,3,"4-10"], if an interval is input, both the lower and the upper bond are included.
     }

   • Optional-ToDos:
     • pipeline configurations:
       ⚠️ We recommand not adjusting the EVCouplings related configs (EVC, unless you are familiar with the parameters) and empirically pre-defined glycan chain topological parameters (bond length, angles, and dihedrals).
       ✅ Conservation_threshold, evc_coupling_threshold, and sasa_cutoff can be appropriately adjusted.

       # An example
       
       pipeline_configs = {
        "Cb_interaction_threshold": 6.0,
        "num_neighbors_to_shield": 3,
        "evc_min_sequence_distance": 6,
        "evc_theta": 0.8,
        "evc_num_iterations": 100,
        "evc_lambda_h": 0.01,
        "evc_lambda_J": 0.01,
        "evc_num_cpu": 10,
        "conservation_threshold": {"loop": 0.5, "ss": 0.5}, # The residues on loop region/other secondary structures with conservation value above the given threshold will be discarded from the prefilter. The higher the thresholds, the less strict the filtering.
        "evc_coupling_threshold": 0.5, # The residues with co-evolving strength above the given threshold will be discarded from the prefilter. The higher the threshold, the less strict the filtering.
        "sasa_cutoff": 0.5, # The residues with SASA values falling in the last n% will be discarded from the prefilter. The higher the threshold, the more strict the filtering.
        "bond_length_C1_ND2": 1.43,
        "angle_C1_ND2_CG": 120.0,
        "angle_C2_C1_ND2": 109.5,
        "angle_O5_C1_ND2": 109.5,
        "dihedral_C1_ND2_CG_CB": {"loop": -120.0, "helix": -180.0, "sheet": 180.0},
        "dihedral_C2_C1_ND2_CG": {"loop": 120.0, "helix": 140.0, "sheet": 100.0},
        "dihedral_O5_C1_ND2_CG": {"loop": -120.0, "helix": -100.0, "sheet": -140.0},
       }

     • ranker configurations:
       ➡️ Manually set the weights for all the factors for glycosylation sites ranking. The higher the value, the more important the factor is.

       # An example
       
       ranker_configs = {
         "conservation_weight": 1.0, # Conservation level. The next 2 values are computed by EVCouplings.
         "coupling_weight": 0.5, # Co-evolving strength.
         "sasa_weight": 1.0, # SASA of the current site. The next 5 values are computed by PyRosetta.
         "sasa_next1_weight": 0.7, # SASA of the next site to the current site.
         "sasa_next2_weight": 0.5, # SASA of the next 2nd site to the current site.
         "sasa_around_weight": 1.0, # Mean SASA value of the triad sequon around the current site.
         "sasa_next_weight": 0.6, # Mean SASA value of the triad sequon following the current site.
         "ddG_weight": 0.3, # ΔΔG of the protein mutant (NXX) and WT. The next 6 values are computed by SPIRED.
         "dTm_weight": 0.3, # ΔTm of the protein mutant (NXX) and WT
         "ddG_S_weight": 0.2, # ΔΔG of the protein mutant (NXS) and WT
         "dTm_S_weight": 0.2, # ΔTm of the protein mutant (NXS) and WT
         "ddG_T_weight": 0.2, # ΔΔG of the protein mutant (NXT) and WT
         "dTm_T_weight": 0.2, # ΔTm of the protein mutant (NXT) and WT
         "mut_score_weight": 0.3, # Mutation score of the protein mutant (NXX). The next 3 values are computed by SaProt.
         "mut_score_S_weight": 0.2, # Mutation score of the protein mutant (NXS).
         "mut_score_T_weight": 0.2, # Mutation score of the protein mutant (NXT).
        }

2. Run the pipeline
   Run the pipeline with the following command:

   python run_sugarswitch.py prefilter --input [input_fasta_file] --out_dir [output_dir]

   • ⚠️ The input file in FASTA format should contain at least one sequence. Please note that the chain ID specified in protein_chain_id in the config.py should correspond to the number and order of the chains

     # An example
     # In this case, test_seq_1 will be chain A-C, and test_seq_2 will be chain D
     
     >test_seq_1 copies: 3
     MSSQIRQNYSTDVEAAVNSLVNLYLQASYTYLSLGFYFDR...
     >test_seq_2 copies: 1
     WNPPTFSPALLVVTEGDNATFTCSFSNTSESFVLNWYRMS...

Acknowledgements

1. Hopf, Thomas A., et al. "The EVcouplings Python framework for coevolutionary sequence analysis." Bioinformatics 35.9 (2019): 1582-1584.
2. Chaudhury, Sidhartha, Sergey Lyskov, and Jeffrey J. Gray. "PyRosetta: a script-based interface for implementing molecular modeling algorithms using Rosetta." Bioinformatics 26.5 (2010): 689-691.
3. Chen, Yinghui, et al. "An end-to-end framework for the prediction of protein structure and fitness from single sequence." Nature Communications 15.1 (2024): 7400.
4. Su, Jin, et al. "Saprot: Protein language modeling with structure-aware vocabulary." BioRxiv (2023): 2023-10.

Licence & Citation

License: MIT License - See LICENSE file for details
Citation: If you use SugarSwitch in your research, please cite:

@article{xxx,
  title={title},
  author={authors},
  journal={journal},
  pages={pages},
  year={2025},
  publisher={publisher}
}