Sactibonds

Evaluating protein structure prediction models by how closely they match known sulfur-to-alpha-carbon post-translational modifications.

Paper

Download the paper on bioRxiv here.

Usage

For cross-linking structure prediction evaluation, run:

python structure_eval.py

For bonds sequence location prediction evaluation, run:

python bonds_eval.py

Sources

Sactibond structures for the following peptides:

• Subtilosin A
• Thurincin H
• Thuricin CD alpha
• Thuricin CD beta
• Streptosactin
• Huazacin
• Qmp A
• Skf B
• Rum C

Models evaluated to date

ESMFold:

Retrieved using the ESMFold API: https://esmatlas.com/.

Script usage:

python peptide_folding.py

AlphaFold 3:

Retrieved using the AlphaFold webserver interface: https://alphafoldserver.com/

AlphaFold 2, Boltz-1. RoseTTAFold2, OmegaFold:

Retrieved using the ColabFold notebooks: https://github.com/sokrypton/ColabFold

Boltz-2:

Retrieved using the Tamarind Bio server: https://app.tamarind.bio/app

Citation

If you find this useful, please cite as follows:

Romain Lacombe. Non-canonical crosslinks confound evolutionary protein structure models. arXiv, 2025.