A familiar frustration in molecular modeling is having a sequence—or a protein–ligand system you want to explore—but not yet having a structure to work with. Choosing a prediction service can feel like another problem before the real modeling begins. A useful first step is to decide what information you have and what you need the prediction workflow to accept.
SAMSON’s Biomolecular Structure Prediction extension brings three cloud prediction services together under Home > Predict: AlphaFold-2, Boltz-2, and Chai-1. They do not all use the same input workflow, so matching the service to your starting materials can save time and help you prepare the job correctly.
Start with your inputs
- Have FASTA files? AlphaFold-2 accepts one or more FASTA files. You can choose a model, such as monomer or multimer, and select a database for multiple sequence alignment before starting the prediction.
- Need to describe a protein, nucleic acid, or ligand together? Boltz-2 lets you add protein, DNA, RNA, and ligand inputs. The tutorial describes entering sequences, CCD codes, or ligand SMILES strings, as appropriate.
- Working with a system where spatial relationships matter to your setup? Chai-1 also accepts protein, DNA, RNA, and ligand inputs, and provides options for adding pocket or contact restraints.
This is a practical way to narrow down the choice: begin with the format of your inputs, then check whether you need the extra input options described for a particular service. The available options are not a claim that one method is best for every target. They are a guide to setting up a job that matches the information you have.
What the restraint options mean in practice
For Boltz-2, you can add a modification to a protein, DNA, or RNA sequence by specifying a residue index and a CCD code. For Chai-1, you can add a pocket restraint to specify a distance relationship between a residue and a chain, or a contact restraint between two residues. These controls let you include additional information when preparing a prediction. Before submitting, make sure residue positions and codes reflect the system you intend to model.
A simple pre-submission checklist
- Identify the components in your system: protein, DNA, RNA, and/or ligand.
- Prepare the required input: FASTA files, sequences, CCD codes, or ligand SMILES strings, depending on the service.
- If using AlphaFold-2, decide which model and multiple-sequence-alignment database to select.
- If using Boltz-2, check any modification’s residue index and CCD code.
- If using Chai-1, review any pocket or contact restraints you plan to add.
- Check the expected use of computing credits before starting the cloud job.
Predictions run in the cloud over a secure connection. You can follow results in SAMSON under Interface > Cloud jobs, or through SAMSON Connect > Account > Jobs. When loaded, structures are colored by pLDDT values if those values are available in the result files. Treat that coloring as information to inspect alongside the model—not as a substitute for deciding whether the structure is suitable for your next step.
Costs and compute options differ by service. AlphaFold-2 offers cloud machines with varying performance and costs, including instances with A100 GPUs. The documentation describes a typical Boltz-2 or Chai-1 prediction as costing between 0.5 and 1 computing credit, using a dedicated A100 GPU. You can buy computing credits or contact SAMSON support to request them.
Once a result is available, the next step is to inspect it and decide whether it is appropriate for comparison, docking, simulation, or visualization. For the full walkthrough and service-specific instructions, read the biomolecular structure prediction tutorial.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.