Protein structure prediction can support functional reasoning, mutation design, and experimental planning, but a predicted structure is not an experimentally determined one. Complex state, conformational change, ligands, modifications, and training distribution can affect results. Confidence is not a direct probability of biological correctness.
When this workflow is useful
- Forming a testable hypothesis without an experimental structure
- Examining domains, interfaces, or possible effects of variants
- Preparing candidate conformations for experiments or comparison
A practical sequence
- 01
Define the biological object
Confirm organism, sequence version, isoform, domains, signal peptides, modifications, and expected assembly. A model cannot repair an input representing the wrong state.
- 02
Inspect sequence and known structures
Search homologues, conserved regions, and experimental structures. Record database versions, identifiers, and manual trimming.
- 03
Run and interpret prediction
Distinguish monomer, complex, and local questions; record version and settings. Examine local and global confidence, inter-chain uncertainty, disorder, and alternatives.
- 04
Test against external evidence
Challenge consequential interpretations with mutagenesis, cross-linking, low-resolution measurements, homologous structures, or suitable experiments. Retain raw outputs.
Quality-control questions
- Do sequence, isoform, and assembly match the question?
- Were confidence and disorder inspected by region?
- Are functional claims independently supported?
- Is prediction excluded from unvalidated clinical decisions?