A structure-prediction workflow often starts with a deceptively simple question: which service should I use, and what should I prepare before I start? A sequence may be ready, while a ligand, chemical modification, or interaction constraint still needs to be represented. Choosing the right input route up front can make the prediction process clearer and reduce avoidable setup friction.
SAMSON’s Biomolecular Structure Prediction extension provides access to AlphaFold-2, Boltz-2, and Chai-1 through Home > Predict. The useful distinction is not simply the names of the models: each service presents a different way to define the prediction task.
Start with the system you want to model
If your input is one or more FASTA files and you want to select an AlphaFold model and a multiple-sequence-alignment database, the AlphaFold-2 workflow is the one described in the tutorial. You choose the service, add the FASTA files, select the model—for example, monomer or multimer—and select the database before starting the prediction.
If your system includes proteins, DNA, RNA, or a ligand, Boltz-2 lets you add those components through separate controls. The inputs can include sequences, CCD codes, or ligand SMILES strings. Boltz-2 also supports adding sequence modifications by specifying a residue position and a CCD code. That makes it particularly important to have the relevant component descriptions and modification details ready before opening the prediction form.
Chai-1 also provides controls for adding proteins, DNA, RNA, and ligands, with sequences or SMILES strings as inputs. Its documented restraint options include a pocket restraint—a distance restraint between a residue and a chain—and a contact restraint between two residues. If a known spatial relationship is important to the question you are asking, decide whether one of these restraints is relevant before launching the job.
A short pre-launch checklist
- Identify the components. List the protein, nucleic acid, and ligand components you want included.
- Gather the right representations. Have FASTA files ready for the AlphaFold-2 workflow; for Boltz-2 or Chai-1, prepare the sequences and, where relevant, CCD codes or ligand SMILES strings.
- Check whether details need encoding. For Boltz-2, note any sequence modifications and their residue positions. For Chai-1, consider whether a pocket or contact restraint is appropriate.
- Review the service and resource requirements. AlphaFold-2 offers a choice of cloud machines with different performance and costs, including instances with A100 GPUs. Boltz-2 and Chai-1 calculations use a dedicated cloud A100 GPU and computing credits; the documentation gives a typical cost of 0.5 to 1 credit for each of those services.
- Plan where to review the result. Once submitted, jobs can be viewed in SAMSON under Interface > Cloud jobs or through SAMSON Connect > Account > Jobs.
Prediction is a step in a larger workflow
A predicted structure is a candidate model to inspect—not a substitute for deciding whether it is suitable for your next task. In SAMSON, loaded prediction results are colored according to pLDDT values when those values are available in the files. This can help you notice confidence information in the result, but it does not by itself validate a model for a particular scientific question.
After reviewing the structure, the next step depends on your goal. You might compare candidates, prepare a chosen model for docking or simulation, or simply visualize it. The SAMSON tutorials also cover protein preparation and validation, protein alignment, and GROMACS Wizard workflows.
For a practical Boltz-2 walkthrough—including sharing job results and generating protein–ligand interaction diagrams—the documentation includes a video tutorial.
For the full setup instructions and references for all three services, read the original biomolecular structure prediction tutorial.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.