Step-by-step guide to predicting binding sites on a protein structure with P2Rank on the platform.

Quick binding site prediction with .bloom* platform

P2Rank uses random forest machine learning to predict ligand binding sites from protein 3D structure. This guide walks through predicting pockets on a protein, from entering a PDB ID to interpreting druggability scores.

Step-by-step pocket prediction workflow

Access the P2Rank tool

Navigate to P2Rank in the .bloom* tool selector, or ask Ivy directly:

"Run P2Rank on 6LU7 to find binding pockets"
"Predict binding sites on my ESMFold structure"

Prepare your protein structure

Option A: PDB ID lookup (fastest)

Enter a PDB accession code (e.g., 6LU7 for SARS-CoV-2 main protease, 1M17 for EGFR kinase)
.bloom* fetches the structure from RCSB PDB automatically
All chains in the structure are analyzed together — there's no chain-filter parameter

Option B: Structure upload

Upload a PDB file from your computer
.bloom* validates the structure and displays chain information
Supports standard PDB format with ATOM records

Option C: From previous jobs

Select a structure from a completed ESMFold prediction
Import from AlphaFold database fetches or PDB search results
.bloom* handles format extraction automatically

Configure prediction parameters

Basic settings (recommended for first run):

Model: default (general-purpose, best for experimental structures)
Visualization: Enable PyMOL script generation
Keep all other parameters at defaults

Advanced settings (via gear icon):

Model variant: default or alphafold
Max pockets: Maximum number of pockets to report (1-50, default 10)
Visualization: Enable/disable PyMOL script generation

Submit and monitor the job

Click Predict Binding Sites to start the job
.bloom* shows real-time progress through structure parsing, feature extraction, and classification
Typical completion: 10-30 seconds per protein
Results appear automatically on the canvas

Review and analyze results

Pocket summary:

Ranked list of predicted pockets with probability scores
Druggability classification for each pocket (druggable, intermediate, difficult, undruggable)
Residue lists showing which amino acids line each pocket

3D visualization:

Predicted pockets highlighted on the protein surface in the Mol* viewer
Color-coded by pocket rank (top pocket in green, secondary pockets in yellow/orange)
Toggle pocket visibility to compare different sites

PyMOL script (if enabled):

Downloadable PyMOL script for offline visualization
Pockets rendered as surface representations with custom coloring
Pocket-lining residues shown as sticks

Download options:

CSV: Pocket table with scores, druggability, residues, and centers
PDB: Structure annotated with pocket occupancy in B-factor column
PyMOL script: Interactive visualization for offline analysis
JSON: Full structured results with all pocket metadata

Common pocket prediction scenarios

Pre-docking site identification

Enter the PDB ID of your target protein
Run P2Rank with the default model
Identify the top-scoring pocket as the primary docking target
Feed the pocket information into DiffDock for targeted docking

Novel target assessment

Predict the structure of a novel protein with ESMFold
Run P2Rank with the alphafold model variant on the predicted structure
Assess whether druggable pockets exist on the predicted structure
Use druggability scores to decide whether to pursue the target

Multi-site analysis

Run P2Rank on a protein with multiple known binding sites
Compare druggability scores across all predicted pockets
Identify the most druggable site for small molecule targeting
Look for allosteric sites beyond the primary active site

Wild-type vs. mutant comparison

Run P2Rank on the wild-type structure
Run P2Rank on the mutant structure (or model the mutation)
Compare pocket rankings and druggability scores
Identify how mutations affect binding site properties

Quality control checklist

Before submission

PDB structure is complete with no missing heavy atoms in the binding region
Structure resolution is reasonable (<3.0 angstroms for experimental structures)
Appropriate model variant chosen (default for experimental, alphafold for predicted structures)

After prediction

Top pocket corresponds to known binding site (if available for validation)
Pocket scores are reasonable (>0.5 for high-confidence pockets)
Pocket-lining residues make biological sense
Druggability classification aligns with known drug target properties

Next steps after pocket prediction

Molecular docking:

Run DiffDock to dock small molecules into the top-scoring pocket
Use pocket residue information to set up targeted docking grids
Compare binding modes across multiple predicted pockets

Target assessment:

Feed druggability scores into Target Passport for comprehensive target evaluation
Compare pocket properties across species orthologs for selectivity analysis
Assess whether the target is amenable to small molecule intervention

Structure-based design:

Use pocket geometry and residue composition to guide pharmacophore design
Identify key interactions (hydrogen bond donors/acceptors, hydrophobic regions) within the pocket
Use pocket shape to filter virtual screening results for shape complementarity

P2Rank quickstart