API Services

The following API services are provided by Cyrus:

AI Folding performs protein structure prediction using Openfold, AlphaFold, and ESMFold. Protein structures are optionally relaxed using Rosetta, and a molprobity report is produced for each model.

Antibody HM creates homology models of antibodies using the Cyrus NextGen Antibody structure prediction method.  This method takes as input an antibody heavy and light chain and returns models of the antibody fV fragment.  The Single Chain HM API is a more suitable approach for modeling nanobodies and non-antibody proteins.

BLAST runs the BLAST+ executable with the v5 NCBI databases.

Clean PDB is a tool made available to all customers free of charge for preparing PDB files for use by the other Cyrus API tools.

DDG is a tool for predicting the energetic consequences of single mutations. It approximates the Gibbs free energy change of folding of the native vs. point mutant.

Deimmunizer is a Rosetta based tool for reducing immunogenicity of proteins by reducing predicted MHC Class II epitopes.

Design is a tool for performing Rosetta based protein design using the FastDesign protocol.

Disulfidizer is a tool for searching for disulfides.  Given an input structure it will attempt to sample disulfides that might exist.  This is appropriate for:

• Looking for disulfides which will stabilize the conformation of an input structure

• Searching for places to put disulfides when no prior hypothesis or suggestion is at hand

Disulfidizer results are reasonably polished (if the –permissive flag is not used), but you may wish to further evaluate disulfides with other computational tools before proceeding to a wet experiment.

Epitope Scan runs the Rosetta MHC II epitope prediction algorithm, as well as the NetMHCPan and NetMHCIIPan prediction algorithms on an input protein structure or sequence. This API uses a machine learning model to predict epitopes based entirely on the sequence of the protein.  Structure input is provided as a convenience but the API will produce identical results regardless of whether a sequence or structure is used as input.

Force Disulfide forces the input disulfide to exist and attempts to relax the protein structure in the presence of that chemical bond (simulated as bond-like restraint, not a kinematic imperative).  It requires the user to state which disulfides are desired.  This tool is appropriate in three cases:

1. You know this bond forms in nature for this sequence (perhaps from mutational data), have a structure of the sequence without cysteines or disulfides, and want a quick peek at what it might look like

2. You are doing blue-sky thinking or hypothesis generation about places disulfides could go and want to absolutely force models to have a given disulfide (our other tools will not enforce it so strongly if the energetics dislike it)

3. You are creating input models for other tools like HM.

Force disulfide results should always be considered preliminary and evaluated with/after other modeling tools.

Glyocosylation prediction predicts glycosylation sites using a Cyrus-developed version of DeepNGlyPred (DNGP)-- a deep neural-network (DNN) learning tool for sequence-based human N-linked glycosylation prediction.

Interface Analyzer runs the Rosetta InterfaceAnalyzer application to report calculated binding energies and other metrics of a given protein-protein interface.

Loop modeling runs the Rosetta Next Generation KIC loop modeling protocol.  This API is useful for modeling loops of between 5 and 17 residues.  This tool is suitable for resampling the conformations of existing loops.  It is the same underlying tool as our Loop Rebuild Action in Bench; that tool’s documentation contains more details.

Make Fragments creates Rosetta fragment libraries for use in Rosetta structure prediction and modeling protocols.

Molprobity generates HTML reports using the MolProbity software developed at Duke University (Main page - MolProbity )

Post-Translational Modification Prediction quickly predicts glycosylation, isomerization, Asn deamidation, and Met photooxidation sites to identify manufacturing liabilities in proteins.

ProteinMPNN uses an AI/ML model to rapidly design sequences which fold into a specified protein backbone

Relax runs the Rosetta FastRelax protocol.  This is useful for taking a structure not generated by Rosetta and “relaxing” it into the Rosetta score function preparatory to further modeling, or for generating a backbone ensemble for a further modeling experiment.

RfDiffusion Runs the Rosetta RFDiffusion AI protein design protocol.

RosettaHoles runs the RosettaHoles application which produces statistics and visualization of voids in proteins. This is a useful tool for evaluating protein designs.

Single Chain HM is a homology modeling tool for generating a structure prediction for your sequence optimizing structures. This tool adds an additional feature to tell Rosetta where extra disulfides are wanted by the researcher, to cover cases where the templates were not clear.  This is appropriate for:

• Creating high-quality models of proteins that you have preliminarily modeled by other means, or have experimental reason to believe they exist

Results from Single Chain HM should be considered gold standard results from Rosetta.  Further modeling never hurts but is not strictly necessary.

Solubility Scoring predicts the solubility of a protein structure using a method similar to the AGGRESCAN method. This is a structure based method and requires a protein model as input, a high quality homology model generated by the Single Chain HM method is an acceptable input to the Solubility Scoring tool.

Template Predictor runs blast, sparksX, and hhsearch, and returns alignment information which can be useful in determining how well homology modeling will work for the given target.

Tolerance Identification finds the closest sequence fragments in the human genome which are present in your protein of interest, this tool is used to screen protein designs for immune tolerance.