PDB formats, model selection, and visualization options for P2Rank pocket prediction.

You provide a protein 3D structure — a PDB ID, an uploaded file, or a structure from a previous job. The one choice that most affects the result is the model variant: use default for experimental crystal structures and alphafold for ESMFold or AlphaFold predictions, so pocket scoring accounts for prediction uncertainty.

Webapp input methods

The .bloom* webapp provides several input methods for P2Rank pocket prediction.

Input type selection

Input Type	Description	Example
PDB ID or UniProt ID (`uniprot_id`)	A single field that accepts either — the tool auto-detects a 4-character PDB ID vs. a UniProt accession and fetches the structure	`6LU7`, `1M17`, `4HJO`, or `P00533`
PDB content (`pdb_content`)	Paste or upload PDB file content directly	Upload `protein.pdb`
From previous job	Import from ESMFold, AlphaFold, or PDB search — carried into `pdb_content` or `uniprot_id` the same way	Select from completed jobs

There's no separate S3-URI parameter — a stored structure from a previous job is read and passed as pdb_content.

Structure requirements

PDB format

ATOM records: Standard PDB format with all heavy atom coordinates
Chain identifiers: Multi-chain structures must have unique chain IDs (A, B, C, etc.)
Standard residues: 20 canonical amino acids; non-standard residues (modified AA, ligands) are ignored
Coordinates: All coordinates in angstroms with reasonable B-factors

Quality requirements

Resolution: Best results with <3.0 angstrom resolution; >4.0 angstrom resolution may produce unreliable pocket predictions
Completeness: Missing heavy atoms in the binding region reduce prediction accuracy; rebuild missing side chains if possible
Alternate conformations: Only the A conformation is used; remove alternate conformers for clarity
Water molecules: Crystallographic waters are ignored; P2Rank uses the protein surface only

Structure preprocessing

Ligand removal: Existing ligands in the PDB are automatically removed before prediction (P2Rank predicts from apo structure)
Hydrogen removal: Hydrogen atoms are not used and are removed if present
Multi-model PDB: Only the first MODEL record is used (relevant for NMR ensembles)
HETATM records: Non-standard residues and small molecules are excluded from the analysis

Model variant selection

Model	Description	Training Data	Best For
default	Standard random forest model	Experimental PDB structures with bound ligands	General-purpose pocket prediction on experimental structures
alphafold	Adapted for predicted structures	AlphaFold and ESMFold predictions	Structures from ESMFold, AlphaFold, or other prediction methods

Choosing a model variant

default: Use for experimental crystal structures from the PDB; best general performance
alphafold: Use for ESMFold or AlphaFold predicted structures; accounts for lower accuracy of predicted models, especially in loop regions

Prediction parameters

These are the actual parameters this tool accepts:

Parameter	Type	Default	Description
uniprot_id	string	-	PDB ID (4-char) or UniProt accession — auto-detected. Provide this OR `pdb_content`
pdb_content	string	-	PDB file content, or a structure Ivy already fetched from the PDB in this chat
model	string	`default`	Model variant (`default`, `alphafold`)
max_pockets	integer	10	Maximum number of pockets to report (1-50)
visualization	boolean	true	Generate PyMOL visualization script

There's no chain-filtering parameter, no minimum pocket score/size threshold, and no color-scheme or surface-type option — P2Rank analyzes the full structure as submitted, and druggability/rank are read from the returned pocket table rather than set as an input filter.

Previous job integration

Compatible tools

ESMFold: Predicted protein structures (use alphafold model variant)
AlphaFold DB: Pre-computed structure predictions
PDB Search: Experimental structures from RCSB PDB
PDB2PQR: Protonated and charge-assigned structures

Data handling

Automatic PDB extraction: .bloom* extracts PDB content from various job output formats
Chain identification: Chain information is parsed and displayed for reference (there's no chain-filter parameter — all chains in the submitted structure are analyzed together)
Provenance tracking: Maintains connection to the original structure source

P2Rank inputs and parameters