This page explains the full lifecycle of a structure in 3decision — from upload to a fully searchable, browsable entry. It covers what you can register, how the registration works, what analyses run automatically after registration, and what actions you can take afterwards.
Registering a structure involves uploading a structure file (.pdb, .cif, .ent, or their .gz variants) and providing metadata. The key steps are:
Public PDB structures from the RCSB are automatically registered in the database on a weekly basis — you do not need to register them yourself.
For step-by-step instructions:
Public AlphaFold models can be registered directly from the AlphaFold Protein Structure Database using a dedicated workflow. See the AlphaFold models feature page and API endpoint details.
Once a structure is registered, 3decision automatically runs several analyses. These analyses make the structure fully searchable and browsable — no manual action is needed.
Each protein chain is mapped to its UniProt reference sequence. For public structures, this mapping is automatically extracted from the structure file header (DBREF records in PDB files or _struct_ref_seq in mmCIF files). For proprietary structures, the same information is used when present in the file header; if it is missing, the chain mapping provided in the registration payload is used instead. Any residues not covered by either source are assigned to a generic sequence (PRO_XXX).
A minimum 80% sequence similarity is required for a successful match. If a structure contains too many mutations to match the canonical UniProt sequence, or if it corresponds to a proprietary synthetic sequence, you can first register a custom sequence and then map the chain to it during structure registration or by re‑running the protein chain sequence mapping analysis.
This mapping:
All small molecules in the structure file are detected and their bond orders are corrected to ensure accurate chemistry representation. The correction uses a priority-based template matching process:
Correct ligand chemistry is essential: it ensures ligands are found through chemistry searches, displayed accurately in 2D and 3D, and classified correctly as drug-like or not. If a ligand cannot be corrected, it will not be registered and the structure is annotated with a missing ligand flag.
If you discover a ligand error after registration, you can fix it without re-uploading — see Ligand Correction.
The protein surface is analyzed using fpocket to identify binding sites:
For each pocket, additional analyses characterize ligand–cavity overlap and physicochemical features (HBA, HBD, charge, aromaticity). Results are available in the Pocket Browser and enable Pocket Similarity Searches.
For structures with more than 50,000 atoms, only explicit pockets (ligand-defined) are detected.
Protein–ligand interactions are detected, classified, and stored. They are:
For structures with more than 50,000 atoms, interactions are not detected.
For technical details on each analysis and how to re-launch them via the API, see the Post-Registration Structure Analyses API documentation.
Once a structure is registered and processed, several actions are available:
| Action | Description | Guide |
|---|---|---|
| Edit metadata | Update title, label, resolution, experimental method | API guide — Tutorial |
| Correct ligands | Fix ligand bond orders or chemistry without re-uploading | API guide — Tutorial |
| Add to project | Assign the structure to one or more projects | Projects |
| Adjust permissions | Manage privileges and user group access | Privileges |
| Add annotations | Attach structure-level or residue-level annotations | Structure annotations — Residue annotations |
| Attach files | Upload additional files (electron density maps, documents, etc.) | API guide — Electron density maps |
| Export structures | Download structures as PDB, mmCIF, or original files | Tutorial |
| Re-run analyses | Re-launch any post-registration analysis via the API | API guide |