A structure-prediction run can stall before it even starts: Which service fits the system? What should you prepare as input? And where will the results appear? The answers depend less on finding one universal option and more on matching your available inputs and modeling needs to a prediction workflow.
SAMSON brings AlphaFold-2, Boltz-2, and Chai-1 together under Home > Predict. Here is a practical way to prepare and choose among them without treating the services as interchangeable.
Start with the inputs you have
If your starting point is one or more FASTA files, the documented AlphaFold-2 workflow is direct: select the service, provide the files, choose a model—such as monomer or multimer—and select a multiple-sequence-alignment database. Then start the prediction. This workflow also lets you choose among cloud machines with different performance and costs; the documentation notes that powerful A100 GPU instances are available.
If your system includes components beyond protein sequences, Boltz-2 offers fields for proteins, DNA, RNA, and ligands. Enter the corresponding sequences, CCD codes, or ligand SMILES strings. You can also specify a sequence modification by residue index and CCD code. This is useful when your prepared input needs to represent a modified protein or nucleic-acid sequence rather than only its unmodified form.
Chai-1 also accepts protein, DNA, RNA, and ligand inputs, with sequences or ligand SMILES strings. Its documented restraint options let you add a pocket restraint between a residue and a chain, or a contact restraint between two residues. If you already know a spatial relationship you want to encode, these options are worth considering when setting up the job.
A small preflight checklist
- Identify the system: protein only, a complex with nucleic acids, a ligand-containing system, or a sequence with modifications.
- Gather the right input: FASTA files for the AlphaFold-2 workflow; sequences, CCD codes, or SMILES strings as appropriate for Boltz-2; and sequences or SMILES strings for Chai-1.
- Decide whether extra setup applies: select an AlphaFold model and alignment database, enter Boltz-2 modifications if needed, or add Chai-1 restraints if you have relationships to specify.
- Check the credit model: the documentation says typical Boltz-2 and Chai-1 predictions cost between 0.5 and 1 computing credit each. AlphaFold-2 uses a choice of cloud machines with varying performance and costs. Check your available credits before launching.
This checklist does not tell you which prediction will be scientifically best for a particular target. It helps prevent a more basic problem: choosing a workflow that does not match the inputs or setup you intended to provide.
Know where results go
Predictions run in the cloud, with data sent through a secure connection. Once a job is submitted, you can follow it in Interface > Cloud jobs in SAMSON or on SAMSON Connect > Account > Jobs. When results are loaded into SAMSON, structures are colorized by pLDDT values when those values are available in the files. That coloring can help you inspect confidence information, but it is not a substitute for evaluating whether a model is suitable for your scientific question.
After prediction, inspect the structures and compare candidates where appropriate. The next step may be visualization, docking, or preparing a selected model for simulation; the right choice depends on your project and the model you obtained.
For the complete service instructions and related workflows, see the SAMSON biomolecular structure prediction tutorial.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.