A structure-prediction job can stall before it even starts—not because the science is difficult, but because the input is in the wrong form or the chosen service does not match the question. You may have a FASTA file, a ligand SMILES string, a modified residue, or a specific interaction you want to constrain. Choosing a prediction route around those inputs can make the setup clearer and help you spend credits deliberately.
SAMSON’s Biomolecular Structure Prediction extension provides access to AlphaFold-2, Boltz-2, and Chai-1 through Home > Predict. Rather than treating them as interchangeable buttons, begin by asking: What molecular components do I need to include, and what information do I already have?
Match the service to your input
- AlphaFold-2: A straightforward option when your inputs are one or more FASTA files. You also choose an AlphaFold model—such as monomer or multimer—and a database for multiple sequence alignment. This makes the model and database choices part of the setup, not something to leave until after the job begins.
- Boltz-2: Useful when you want to assemble a prediction from protein, DNA, RNA, or ligand components. The interface accepts the corresponding sequences, CCD codes, or ligand SMILES strings. For protein, DNA, or RNA modifications, you can provide a residue position and the modification’s CCD code.
- Chai-1: Also lets you add protein, DNA, RNA, and ligand inputs. If you have a specific spatial relationship to express, its interface offers pocket restraints between a residue and a chain, and contact restraints between two residues.
This is a practical distinction: first identify whether your job is mainly a FASTA-based prediction, a multi-component input, or a prediction where you want to specify distance relationships. The available inputs and controls can help narrow down which service to try. They do not, by themselves, guarantee that a particular prediction will be correct.
A short preflight checklist
- Gather the inputs. Have your sequences, FASTA files, CCD codes, or SMILES strings ready before opening the prediction interface.
- Check component identity and order. For a multi-component job, make sure you know which sequences or ligand representation belong in each input field.
- Decide whether you need a constraint. If using Chai-1, prepare the residue and chain relationship for a pocket restraint, or the two residues for a contact restraint. If using Boltz-2 modifications, note the residue index and CCD code.
- Consider the cost before submitting. Boltz-2 and Chai-1 calculations use computing credits on a dedicated cloud A100 GPU; a typical prediction costs between 0.5 and 1 credit. AlphaFold-2 uses cloud machines with different performance and costs, including A100 GPU instances, so check the available options for your job.
- Know where results will appear. Results can be viewed in SAMSON under Interface > Cloud jobs or on SAMSON Connect > Account > Jobs.
After the prediction: inspect, don’t just download
When results are loaded into SAMSON, structures are colorized according to pLDDT values if those values are included in the files. Treat this as a useful aid for inspection, not as a substitute for evaluating whether the model answers your biological question. Compare candidates when appropriate, inspect the relevant regions, and decide what additional validation or preparation is needed before docking or simulation. A predicted structure is a model to assess—not an experimental result.
Cloud predictions are sent over a secure connection. For AlphaFold-2, publications reporting findings from use of the service should cite the AlphaFold paper and, where applicable, the AlphaFold-Multimer paper. The documentation also lists the Boltz-2 and Chai-1 references.
For the exact setup steps and service details, read the biomolecular structure prediction tutorial.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.