A structure prediction can stall before it starts—not because the science is complicated, but because the input is not ready or the chosen service does not match the system. A protein sequence, a ligand SMILES string, and a protein–ligand complex are different kinds of inputs. Getting clear on what you have before opening the prediction tool can save time and help you set up a more appropriate job.
SAMSON’s Biomolecular Structure Prediction extension provides access to AlphaFold-2, Boltz-2, and Chai-1 through Home > Predict. The useful first step is not picking a model by name; it is describing your system and preparing the information that the selected service asks for.
Start with the system you want to model
If your input is one or more FASTA files and you want to choose an AlphaFold model—such as monomer or multimer—AlphaFold-2 is the workflow described for that setup. You can also choose the database used for multiple sequence alignment. Have the FASTA files ready, then select the service, model, and database before starting the prediction.
If the system includes several molecular types, Boltz-2 or Chai-1 may fit the input workflow more naturally. Both let you add proteins, DNA, RNA, or ligands. For Boltz-2, protein, DNA, and RNA inputs can include modifications specified by a residue index and a CCD code. Ligands can be entered as CCD codes or SMILES strings. Chai-1 accepts sequences for proteins, DNA, and RNA, and SMILES strings for ligands; it also provides options for adding restraints.
Use restraints only when they reflect information you have
Chai-1 offers pocket restraints and contact restraints. A pocket restraint specifies a distance relationship between a residue and a chain; a contact restraint specifies one between two residues. These options are useful when you have a reason to encode a spatial relationship in the prediction setup. If you do not have such information, you can leave them out rather than inventing constraints.
For Boltz-2, sequence modifications are entered explicitly: choose Add modification, provide the residue position, and enter the modification’s CCD code. Preparing these details ahead of time makes it easier to check that the intended modification and position are represented in the job.
Check the practical details before launching
- Gather the inputs: FASTA files, molecular sequences, CCD codes, or SMILES strings, depending on the selected service.
- Choose the matching workflow: AlphaFold-2 uses FASTA files and offers model and multiple-sequence-alignment database choices; Boltz-2 and Chai-1 provide ways to assemble inputs involving different molecular types.
- Review costs and credits: Boltz-2 and Chai-1 run on a dedicated cloud A100 GPU and use computing credits. A typical prediction with either is described as costing 0.5–1 credit. AlphaFold-2 offers cloud machines with varying performance and costs, including A100 instances.
- Remember where results appear: You can follow jobs in SAMSON under Interface > Cloud jobs or through SAMSON Connect under Account > Jobs.
Predictions run in the cloud, and data are sent via a secure connection. When results are loaded into SAMSON, structures are colorized using pLDDT values if those values are available in the files. Treat that coloring as information to inspect alongside the structure, not as a substitute for evaluating whether the model is suitable for your next step. The documentation also points to protein preparation and validation, alignment, and simulation tutorials for downstream work.
A simple routine helps: identify the molecular components, collect the exact input formats, decide whether modifications or restraints are supported and relevant, check the credit requirements, then launch and inspect the result. For the full step-by-step instructions and service details, read the SAMSON structure prediction tutorial.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.